Unit I Frame relay Networks Frame Relay often is described as a streamlined version of X.25, offering fewer of the robust capabilities, such as windowing and retransmission of last data that are offered in X.25. Frame Relay Devices Devices attached to a Frame Relay W! W! fall into the following two general categories" #
Data Data term termin inal al e$ui e$uipm pmen entt %D& %D&'( '( #
Data Data cir circu cuit it)t )ter ermi minat natin ing g e$ui e$uipm pmen entt %D* %D*'( '(
D&'s generally are considered to be terminating e$uipment for a specific networ+ and typically are located on the premises of a customer. n fact, they may be owned by the customer. customer. '-amples '-amples of D&' devices devices are terminals, terminals, personal personal computers, computers, routers, and bridges. D*'s are carrier)owned internetwor+ing devices. &he purpose of D*' e$uipment is to provide cloc+ing and switching services in a networ+, which are the devices that actually a ctually transmit data through the W!. n most cases, these are pac+et switches. Figure /) shows the relationship between the two categories of devices.
Standard Frame Relay Frame 0tandard Frame Relay frames consist of the fields illustrated in Figure /)1. Figure Five Fields *omprise the Frame Relay Frame
'ach frame relay D3 D3 consists consists of the following fields" . Flag Field. Field. &he &he flag is used to perfor perform m high level level data lin+ lin+ synchroni synchroni4ation 4ation which which indicates the beginning and end of the frame with the uni$ue pattern / //. /. &o ensure that the // pattern does not appear somewhere inside the frame, bit stuffing and destuffing procedures are used. 2. ddress ddress Field. Field. 'ach address address field field may occupy occupy either either octet 2 to , octet octet 2 to 1, or octet 2 to 5, depending on the range of the address in use. two)octet address
fiel field d comp compri risi sing ng the the '6 '6DDR DDR'0 '00 0 F'7 F'7D D 'X&' 'X&'!0 !08 8! ! 9&0 9&0 and the the *:R6*8;;!D:R'08!0' 9&. . D7*)D D7*)Data ata 7in+ *onnecti *onnection on denti dentifie fierr 9its. 9its. &he D7* serves serves to identi identify fy the virtual connection so that the receiving end +nows which information connection a frame belongs to. !ote that this D7* has only local significance. single physical channel can multiplemultiple- several several different virtual connections. 1. F'*!, 9'*!, D' bits. bits. &hese &hese bits bits report report congestion" congestion" F'*!6Forward '-plicit *ongestion !otification bit o 9'*!69ac+ward '-plicit *ongestion !otification bit o D'6Discard 'ligibility bit o 5. nform nformati ation on Field. syste system m parame parameter ter defines defines the ma-imum ma-imum number number of data data bytes that a host can pac+ into a frame.
Congestion-Control Mechanisms Frame Relay reduces networ+ overhead by implementing simple congestion)notification mechanisms rather than e-plicit, per)virtual)circuit flow control. Frame Relay typically is implemented implemented on reliable reliable networ+ media, so data integrity integrity is not sacrifice sacrificed d because flow control can be left to higher)layer protocols. Frame Relay implements two congestion) notification mechanisms" #
Forwar Forward)e d)e-pl -plici icitt cong congest estion ion notifi notificati cation on %F'*!( %F'*!(
# 9ac+ 9ac+wa ward rd)e )e-p -pli lici citt cong conges esti tion on noti notifi fica cati tion on %9'* %9'*!( !( F'*! F'*! and and 9'*! 9'*! each each is controlled by a single bit contained in the Frame Relay frame header. &he Frame Relay frame header also contains a Discard 'ligibility %D'( bit, which is used to identify less important traffic that can be dropped during periods of congestion.
Frame Relay versus X.25 &he design of X.25 aimed to provide error)free delivery over lin+s with high error)rates. Frame relay ta+es advantage of the new lin+s with lower error)rates, enabling it to eliminate many of the services provided by X.25. &he elimination of functions and fields, combined with digital lin+s, enables frame relay to operate at speeds 2/ times greater than X.25.
2
fiel field d comp compri risi sing ng the the '6 '6DDR DDR'0 '00 0 F'7 F'7D D 'X&' 'X&'!0 !08 8! ! 9&0 9&0 and the the *:R6*8;;!D:R'08!0' 9&. . D7*)D D7*)Data ata 7in+ *onnecti *onnection on denti dentifie fierr 9its. 9its. &he D7* serves serves to identi identify fy the virtual connection so that the receiving end +nows which information connection a frame belongs to. !ote that this D7* has only local significance. single physical channel can multiplemultiple- several several different virtual connections. 1. F'*!, 9'*!, D' bits. bits. &hese &hese bits bits report report congestion" congestion" F'*!6Forward '-plicit *ongestion !otification bit o 9'*!69ac+ward '-plicit *ongestion !otification bit o D'6Discard 'ligibility bit o 5. nform nformati ation on Field. syste system m parame parameter ter defines defines the ma-imum ma-imum number number of data data bytes that a host can pac+ into a frame.
Congestion-Control Mechanisms Frame Relay reduces networ+ overhead by implementing simple congestion)notification mechanisms rather than e-plicit, per)virtual)circuit flow control. Frame Relay typically is implemented implemented on reliable reliable networ+ media, so data integrity integrity is not sacrifice sacrificed d because flow control can be left to higher)layer protocols. Frame Relay implements two congestion) notification mechanisms" #
Forwar Forward)e d)e-pl -plici icitt cong congest estion ion notifi notificati cation on %F'*!( %F'*!(
# 9ac+ 9ac+wa ward rd)e )e-p -pli lici citt cong conges esti tion on noti notifi fica cati tion on %9'* %9'*!( !( F'*! F'*! and and 9'*! 9'*! each each is controlled by a single bit contained in the Frame Relay frame header. &he Frame Relay frame header also contains a Discard 'ligibility %D'( bit, which is used to identify less important traffic that can be dropped during periods of congestion.
Frame Relay versus X.25 &he design of X.25 aimed to provide error)free delivery over lin+s with high error)rates. Frame relay ta+es advantage of the new lin+s with lower error)rates, enabling it to eliminate many of the services provided by X.25. &he elimination of functions and fields, combined with digital lin+s, enables frame relay to operate at speeds 2/ times greater than X.25.
2
X.25 specifies processing at layers , 2 and of the 80 model, while frame relay operate operatess at layers layers and 2 only only. &his &his means means that that frame frame relay relay has signific significant antly ly less less processing to do at each node, which improves throughput by an order of magnitude. X.25 prepares and sends pac+ets, while frame relay prepares and sends frames. X.25 pac+ets contain co ntain several fields used for error and flow control, none of which frame relay needs. &he frames in frame relay contain an e-panded address field that enables frame relay nodes to direct frames to their destinations with minimal processing . X.25 has a fi-ed bandwidth available. t uses or wastes portions of its bandwidth as the load load dict dictat ates es.. Fram Framee rela relay y can can dynam dynamic ical ally ly allo allocat catee band bandwi widt dth h durin during g call call setu setup p negotiation at both the physical and logical channel level.
Asynchronous Transfer Transfer Mode Mode (ATM) (ATM) Asynchronous Transfer Transfer Mode (ATM) is an nternatio nternational nal &elecommu elecommunicati nication on 3nion) &elecom lecommu muni nicat catio ions ns 0tan 0tanda dard rdss 0ect 0ectio ion n %&3 %&3)& )&(( stan standa dard rd for for cell cell rela relay y wher wherei ein n information for multiple service types, such as voice, video, or data, is conveyed in small, fi-ed)si4e cells. &; &; networ+s are connection)oriented. &; is a cell)swi cell)switchin tching g and multip multiple-ing le-ing technology technology that that combines combines the benefit benefitss of circuit switching %guaranteed capacity and constant transmission delay( with those of pac+et switching %fle-ibility and efficiency for intermittent traffic(. t provides scalable bandwidth from a few megabits per second %;bps( to many gigabits per second %>bps(. 9ecau 9ecause se of its its asynch asynchro ronou nouss natu nature re,, &; is more more effi effici cien entt than than synch synchro rono nous us technologies, such as time-division multiplexing (TDM). (TDM). With With &D;, each user is assigned to a time slot, and no other station can send in that time slot. f a station has much data to send, it can send only when its time slot comes up, even if all other time slots are empty.
ATM Protocol architecture:
&; is almost similar to cell relay and pac+ets witching using X.25and framerelay.li+e pac+et switching and frame relay,&; involves the transfer of data in discrete pieces.also,li+e pac+et switching and frame relay ,&; allows multiple logical connections to multiple-ed over a single physical interface. in the case of &;,the information flow on each logical connection is organised into fi-ed)si4e pac+ets, called cells. &; is a streamlined protocol with minimal error and flow control capabilities "this reduces the overhead of processing &; cells and reduces the number of overhead bits re$uired with each cell, thus enabling &; to operate at high data rates.the use of fi-ed) si4e cells simplifies the processing re$uired at each &; node,again supporting the use of &; at high data rates. &he &; architecture uses a logical model to describe the functionality that it supports. &; functionality corresponds to the physical layer and part of the data lin+ layer of the 80 reference model. . the protocol referencce model shown ma+es reference to three separate planes" user plane provides for user information transfer ,along with associated controls %e.g.,flow control ,error control(. control plane performs call control and connection control functions. management plane includes plane management ,which performs management function related to a system as a whole and provides coordination between all the planes ,and layer management which performs management functions relating to resource and parameters residing in its protocol entities .
&he &; reference model is composed of the following &; layers" # Physical layer @nalogous to the physical layer of the 80 reference model, the &; physical layer manages the medium)dependent transmission. # ATM layer @*ombined with the &; adaptation layer, the &; layer is roughly analogous to the data lin+ layer of the 80 reference model. &he &; layer is responsible for the simultaneous sharing of virtual circuits over a physical lin+ %cell multiple-ing( and passing cells through the &; networ+ %cell relay(. &o do this, it uses the A and A* information in the header of each &; cell. # ATM adaptation layer (AAL) @*ombined with the &; layer, the 7 is roughly analogous to the data lin+ layer of the 80 model. &he 7 is responsible for isolating higher)layer protocols from the details of the &; processes. &he adaptation layer prepares user data for conversion into cells and segments the data into 1?)byte cell payloads. Finally, the higher layers residing above the 7 accept user data, arrange it into pac+ets, and hand it to the 7. Figure "illustrates the &; reference model.
1
Structure of an ATM cell n &; cell consists of a 5 byte header and a 1? byte payload. &he payload si4e of 1? bytes was a compromise between the needs of voice telephony and pac+et networ+s, obtained by a simple averaging of the 30 proposal of =1 bytes and 'uropean proposal of 2, said by some to be motivated by a 'uropean desire not to need echo)cancellers on national trun+s. &; defines two different cell formats" !! %!etwor+)networ+ interface( and 3! %3ser)networ+ interface(. ;ost &; lin+s use 3! cell format. iagram o! the UNI ATM "ell
iagram o! the NNI ATM "ell
B
B
1
/
>F*
A
A
A
A*
A
A* A*
/
A*
A* &
*7
<'*
ayload
1
A*
&
*7
<'*
%1?
bytes(
ayload
%1?
bytes(
5
>F* 6 >eneric Flow *ontrol %1 bits( %default" 1)4ero bits( A 6 Airtual ath dentifier %? bits 3!( or %2 bits !!( A* 6 Airtual channel identifier %= bits( & 6 ayload &ype % bits( *7 6 *ell 7oss riority %)bit( <'* 6
F* field is reserved for a local flow control:submultiple-ing system between users. &his was intended to allow several terminals to share a single networ+ connection, in the same way that two 0D! phones can share a single basic rate 0D! connection. ll four >F* bits must be 4ero by default.&he !! cell format is almost identical to the 3! format, e-cept that the 1)bit >F* field is re)allocated to the A field, e-tending the A to 2 bits. &hus, a single !! &; interconnection is capable of addressing almost 22 As of up to almost 2= A*s each %in practice some of the A and A* numbers are reserved(.
#irtual "hannel (#") denotes the transport of &; cells which have the same uni$ue identifier, called the Airtual *hannel dentifier %A*(. &his identifier is encoded in the cell header. virtual channel represents the basic means of communication between two end)points, and is analogous to an X.25 virtual circuit. #irtual Path (#P) denotes the transport of &; cells belonging to virtual channels which share a common identifier, called the Airtual ath dentifier %A(, which is also
=
encoded in the cell header. virtual path, in other words, is a grouping of virtual channels which connect the same end)points. &his two layer approach results in improved networ+ performance. 8nce a virtual path is set up, the addition:removal of virtual channels is straightforward
ATM Classes of Services ATM is connection oriented and allows the user to specify the resources required on a per-connection basis (per SVC) dynamically. There are the fie classes of serice defined for ATM (as per ATM !orum "#$ %.& specification). The 'oS parameters f or these serice classes are summaried in Table 1.
$er%ice "lass
&uality o! $er%ice Parameter
&his class is used for emulating circuit switching. &he cell rate is constant bit rate constant with time. *9R applications are $uite sensitive to cell)delay %*9R( variation. '-amples of applications that can use *9R are telephone traffic %i.e., n-=1 +bps(, videoconferencing, and television. &his class allows users to send traffic at a rate that varies with time variable bit rateE depending on the availability of user information. 0tatistical non)real time multiple-ing is provided to ma+e optimum use of networ+ resources. %A9RE!R&( ;ultimedia e)mail is an e-ample of A9RE!R&. &his class is similar to A9RE!R& but is designed for applications that variable bit rateE are sensitive to cell)delay variation. '-amples for real)time A9R are real time %A9RE voice with speech activity detection %0D( and interactive compressed R&( video. &his class of &; services provides rate)based flow control and is aimed at data traffic such as file transfer and e)mail. lthough the standard does not re$uire the cell transfer delay and cell)loss ratio to be available bit rate guaranteed or minimi4ed, it is desirable for switches to minimi4e delay %9R( and loss as much as possible. Depending upon the state of congestion in the networ+, the source is re$uired to control its rate. &he users are allowed to declare a minimum cell rate, which is guaranteed to the connection by the networ+. unspecified rate %39R(
bit &his class is the catch)all, other class and is widely used today for &*:.
Technical Parameter
e!inition
cell loss ratio *7R is the percentage of cells not delivered at their destination
B
%*7R(
because they were lost in the networ+ due to congestion and buffer overflow.
&he delay e-perienced by a cell between networ+ entry and e-it cell transfer points is called the *&D. t includes propagation delays, delay %*&D( $ueuing delays at various intermediate switches, and service times at $ueuing points. cell delay *DA is a measure of the variance of the cell transfer delay. variation
'ene!its o! ATM
&he benefits of &; are the following" •
high performance via hardware switching
•
dynamic bandwidth for bursty traffic class)of)service support for multimedia scalability in speed and networ+ si4e common 7!:W! architecture opportunities for simplification via A* architecture international standards compliance
• • • • •
ATM Adaptation Layers (AAL) &he use of synchronous &ransfer ;ode %&;( technology and services creates the need for an adaptation layer in order to support information transfer protocols, which are not based on &;. &his adaptation layer defines how to segment and reassemble higher)layer pac+ets into &; cells, and how to handle various transmission aspects in the &; layer. '-amples of services that need adaptations are >igabit 'thernet, , Frame Relay, 08!'&:0D<, 3;&0:Wireless, etc. &he main services provided by 7 %&; daptation 7ayer( are" •
0egmentation and reassembly
?
• • •
&he following &; daptation 7ayer protocols %7s( have been defined by the &3)&. t is meant that these 7s will meet a variety of needs. &he classification is based on whether a timing relationship must be maintained between source and destination, whether the application re$uires a constant bit rate, and whether the transfer is connection oriented or connectionless. •
•
•
•
AAL Type supports constant bit rate %*9R(, synchronous, connection oriented traffic. '-amples include & %D0(, ', and -=1 +bit:s emulation. AAL Type supports time)dependent Aariable 9it Rate %A9R)R&( of connection) oriented, synchronous traffic. '-amples include Aoice over &;. 72 is also widely used in wireless applications due to the capability of multiple-ing voice pac+ets from different users on a single &; connection. AAL Type *+, supports A9R, data traffic, connection)oriented, asynchronous traffic %e.g. X.25 data( or connectionless pac+et data %e.g. 0;D0 traffic( with an additional 1)byte header in the information payload of the cell. '-amples include Frame Relay and X.25. AAL Type - is similar to 7 :1 with a simplified information header scheme. &his 7 assumes that the data is se$uential from the end user and uses the ayload &ype ndicator %&( bit to indicate the last cell in a transmission. '-amples of services that use 7 5 are classic over &;, 'thernet 8ver &;, 0;D0, and 7! 'mulation %7!'(. 7 5 is a widely used &; adaptation layer protocol. &his protocol was intended to provide a streamlined transport facility for higher)layer protocols that are connection oriented.
7 5 was introduced to" • • •
reduce protocol processing overhead. reduce transmission overhead. ensure adaptability to e-isting transport protocols.
& AAL1 PDU &he structure of the 7 D3 is given in the following illustration"
S#
CS$
S#
SC
C*C
+C
SA* ," ayload
1B bytes bit
bits bits
bit
AAL1 PD
S#
0e$uence number. !umbers the stream of 0R D3s of a **0 D3 %modulo =(. &he se$uence number is comprised of the *0 and the 0!. CS$
*onvergence sublayer indicator. 3sed for residual time stamp for cloc+ing. SC
0e$uence count. &he se$uence number for the entire *0 D3, which is generated by the *onvergence 0ublayer. S#
0e$uence number protection. *omprised of the *R* and the '*. C*C
*yclic redundancy chec+ calculated over the 0R header. +C
'ven parity chec+ calculated over the *R*. SA*
,"
payload
1B)byte user information field.
AAL 72 provides bandwidth)efficient transmission of low)rate, short and variable pac+ets in delay sensitive applications. t supports A9R and *9R. 72 also provides for variable payload within cells and across cells. 7 type 2 is subdivided into the *ommon art 0ublayer %*0 ( and the 0ervice 0pecific *onvergence 0ublayer %00*0 (. AAL CPS Pac!et
&he *0 pac+et consists of a octet header followed by a payload. &he structure of the 72 *0 pac+et is shown in the following illustration. *D
7
33
<'*
nformation payload
? bits
= bits
5 bits 5 bits )15:=1 bytes
AAL! "P# pac$et *D *hannelidentification. 7 7ength indicator. &his is the length of the pac+et payload associated with each individual user. Aalue is one less than the pac+et payload and has a default value of 15 bytes %may be set to =1 bytes(. 33 3ser)to)user indication. rovides a lin+ between the *0 and an appropriate 00*0 that satisfies the higher layer application
/
<'*
0!
= bits bit
field
*0)D3 payload 72 D3 payload
bit
D /)1B bytes
AAL! "P# PD 80F 8ffset field. dentifies the location of the start of the ne-t *0 pac+et within the *0) D3. 0! 0e$uence number. rotects data integrity. arity. rotects the start field from errors. 0R nformation field of the 0R D3.
D3
payload
D adding. AAL $$"$ Packet
&he 00*0 conveys narrowband calls consisting of voice, voiceband data or circuit mode data. 00*0 pac+ets are transported as *0 pac+ets over 72 connections. &he *0 pac+et contains a 00*0 payload. &here are 00*0 pac+et types. &ype 3nprotectedG this is used by default. &ype 2 artially protected. &ype Fully protected" the entire payload is protected by a /)bit *R* which is computed as for 8; cells. &he remaining 2 bits of the 2)octet trailer consist of the message type field. AAL $$"$ Type * Packets.
&he type pac+ets are used for the following" •
Dialled digits
• • • •
*hannel associated signalling bits Facsimile demodulated control data larms 3ser state control operations.
&he following illustration gives the general sturcture of 72 00*0 &ype D3s. &he format varies and each message has its own format according to the actual message type. Redundancy
&ime stamp
;essage dependant information
;essage type
*R*)/
2
1
=
=
/ bits
AAL! ##"# Type % PD Redundancy ac+ets are sent times to ensure error correction. &he value in this field signifies the transmission number. &ime stamp *ounters pac+et delay variation and allows a receiver to accurately reproduce the relative timing of successive events separated by a short interval. ;essage dependant ac+et content that varies, depending on the message type. ;essage &he message type code.
information
type
*R*)/ &he /)bit *R*. AAL*+,
7:1 consists of message and streaming modes. t provides for point)to)point and point)to)multipoint %&; layer( connections. &he *onvergence 0ublayer %*0( of the &; daptation 7ayer %7( is divided into two parts" service specific %00*0 ( and common part %**0 (. &his is illustrated in the following diagram" 7:1 pac+ets are used to carry computer data, mainly 0;D0 traffic. AAL*+, "P"$ PU
&he functions of the 7:1 **0 include connectionless networ+ layer %*lass D(, meaning no need for an 00*0G and frame relaying telecommunication service in *lass *. &he **0 D3 is composed of the following fields"
2
nfo
&railer
*
9tag
9asi4e **0 0D3
ad
/
'tag 7ength
2
/)=555 /)
2 bytes
AAL%&' "P"# PD
* ;essage type. 0et to 4ero when the 9si4e and 7ength fields are encoded in bytes. 9tag 9eginning tag. &his is an identifier for the pac+et. t is repeated as the 'tag. 9si4e 9uffer allocation si4e. 0i4e %in bytes( that the receiver has to allocate to capture all the data. **0 Aariable information field up to =555 bytes.
0D3
D adding field which is used to achieve 2)bit alignment of the length of the pac+et. / ll)4ero. 'tag 'nd tag. ;ust be the same as 9tag. 7ength ;ust be the same as 90i4e. AAL*+, $A/ PU
&he structure of the 7:1 0R D3 is illustrated below" 0&
0!
;D
nformation
7
*R*
2
1
/
52
=
/ bits
11 bytes
2)byte trailer
2)byte header 1? bytes
AAL%&' #A PD 0& 0egment type. Aalues may be as follows" 0! 0e$uence number. !umbers the stream of 0R D3s of a **0 D3 %modulo =(. ;D ;ultiple-ing identification. &his is used for multiple-ing several 7:1 connections over one &; lin+. nformation &his field has a fi-ed length of 11 bytes and contains parts of **0 D3. 7 7ength indication. *ontains the length of the 0R 0D3 in bytes, as follows" *R* *yclic redundancy chec+. Functions of 7:1 0R include identification of 0R 0D3sG error indication and handlingG 0R 0D3 se$uence continuityG multiple-ing and demultiple-ing. AAL- &he type 5 adaptation layer is a simplified version of 7:1. t also consists of message and streaming modes, with the *0 divided into the service specific and common part. 75 provides point)to)point and point)to)multipoint %&; layer( connections.
75 is used to carry computer data such as &*:. t is the most popular 7 and is sometimes referred to as 0'7 %simple and easy adaptation layer(. AAL- "P"$ PU
&he 75 **0 D3 is composed of the following fields" nfo
&railer
**0 payload
ad
33 * 7ength *R*
/)=555
/)1B
2
1 bytes
AAL "P"# PD **0 &he actual information that is sent by the user. !ote that the information comes before any length indication %as opposed to 7:1 where the amount of memory re$uired is +nown in advance(.
1
ad adding bytes to ma+e the entire pac+et %including control and *R*( fit into a 1?)byte boundary. 33 **0 user)to)user indication to transfer one byte of user information. * *ommon part indicator is a filling byte %of value /(. &his field is to be used in the future for layer management message indication. 7ength 7ength of the user information without the ad. *R* *R*)2. 3sed to allow identification of corrupted transmission. AAL- $A/ PU &he structure of the 75 *0 D3 is as follows"
nformation
D
33
*
7ength *R*)2
)1?
/)1B
2
1 bytes
?)byte trailer AAL #A PD
0igh1$peed LANs 2mergence o! 0igh1$peed LANs 2 0ignificant trends
E E
*omputing power of *s continues to grow rapidly !etwor+ computing '-amples of re$uirements
E E E
*entrali4ed server farms ower wor+groups
E E
transmission from any station can be received by all stations
0olution to First roblem Data transmitted in bloc+s called frames"
E E
3ser data Frame header containing uni$ue address of destination station
5
*0;:*D *arrier 0ense ;ultiple ccess: *arrier Detection f the medium is idle, transmit. f the medium is busy, continue to listen until the channel is idle, then transmit immediately. f a collision is detected during transmission, immediately cease transmitting. fter a collision, wait a random amount of time, then attempt to transmit again %repeat from step (.
;edium 8ptions at /;bps Hdata rateI Hsignaling methodI Hma- lengthI /9ase5 E / ;bps E 5/)ohm coa-ial cable bus E ;a-imum segment length 5// meters /9ase)& E &wisted pair, ma-imum length // meters
=
E
0tar
topology
%hub
or
multipoint
repeater
at
central
point(
B
9ridge
Frame handling done in software naly4e and forward one frame at a time 0tore)and)forward
7ayer 2 0witch Frame handling done in hardware ;ultiple data paths and can handle multiple frames at a time *an do cut)through 7ayer 2 0witches Flat address space 9roadcast storm 8nly one path between any 2 devices
0olution " subnetwor+s connected by routers 0olution 2" layer switching, pac+et)forwarding logic in hardware
?
9enefits of / >bps 'thernet over &; !o e-pensive, bandwidth consuming conversion between 'thernet pac+ets and &; cells !etwor+ is 'thernet, end to end plus 'thernet offers Jo0 and traffic policing capabilities approach that of &; Wide variety of standard optical interfaces for / >bps 'thernet Fibre *hannel 2 methods of communication with processor"
E E
E E
:8 channel !etwor+ communications Fibre channel combines both 0implicity and speed of channel communications Fle-ibility and interconnectivity of networ+ communications
:8 channel
2/
Fibre *hannel !etwor+)8riented Facilities Full multiple-ing between multiple destinations eer)to)peer connectivity between any pair of ports nternetwor+ing with other connection technologies Fibre *hannel Re$uirements Full duple- lin+s with 2 fibres:lin+ // ;bps E ?// ;bps Distances up to / +m 0mall connectors high)capacity >reater connectivity than e-isting multidrop channels 9road availability 0upport for multiple cost:performance levels 0upport for multiple e-isting interface command sets Fibre *hannel rotocol rchitecture F*)/ hysical ;edia F*) &ransmission rotocol F*)2 Framing rotocol F*) *ommon 0ervices F*)1 ;apping
3ireless LAN /e4uirements
&hroughput !umber of nodes *onnection to bac+bone 0ervice area 9attery power consumption &ransmission robustness and security *ollocated networ+ operation 7icense)free operation
''' ?/2. 0ervices ssociation Reassociation Disassociation uthentication rivacy
2
22
Unit II
Jueing analysis n $ueueing theory, a 4ueueing model is used to appro-imate a real $ueueing situation or system, so the $ueueing behaviour can be analysed mathematically. Jueueing models allow a number of useful steady state performance measures to be determined, including" • • • • •
the average number in the $ueue, or the system, the average time spent in the $ueue, or the system, the statistical distribution of those numbers or times, the probability the $ueue is full, or empty, and the probability of finding the system in a particular state. &hese performance measures are important as issues or problems caused by $ueueing situations are often related to customer dissatisfaction with service or may be the root cause of economic losses in a business. nalysis of the relevant $ueueing models allows the cause of $ueueing issues to be identified and the impact of any changes that might be wanted to be assessed. Notation
&ueueing models can be represented using Kendalls notation"
:9:0:K:!:Disc
• • • • • •
where" is the interarrival time distribution 9 is the service time distribution 0 is the number of servers K is the system capacity ! is the calling population Disc is the service discipline assumed 0ome standard notation for distributions % or 9( are"
• • • • •
; for a ;ar+ovian %e-ponential( distribution 'L for an 'rlang distribution with L phases D for Deterministic %constant( > for >eneral distribution < for a hase)type distribution Models Constr"ction and analysis
Jueueing models are generally constructed to represent the steady state of a $ueueing system, that is, the typical, long run or average state of the system.
2
s a conse$uence, these are stochastic models that represent the probability that a $ueueing system will be found in a particular configuration or state. general procedure for constructing and analysing such $ueueing models is" . dentify the parameters of the system, such as the arrival rate, service time, Jueue capacity, and perhaps draw a diagram of the system. 2. dentify the system states. % state will generally represent the integer number of customers, people, Mobs, calls, messages, etc. in the system and may or may not be limited.( . Draw a state transition diagram that represents the possible system states and identify the rates to enter and leave each state. &his diagram is a representation of a ;ar+ov chain. 1. 9ecause the state transition diagram represents the steady state situation between state there is a balanced flow between states so the probabilities of being in adMacent states can be related mathematically in terms of the arrival and service rates and state probabilities. 5. '-press all the state probabilities in terms of the empty state probability, using the inter)state transition relationships. =. Determine the empty state probability by using the fact that all state probabilities always sum to . Whereas specific problems that have small finite state models are often able to be analysed numerically, analysis of more general models, using calculus, yields useful formulae that can be applied to whole classes of problems.
Single-server #"e"e 0ingle)server $ueues are, perhaps, the most commonly encountered $ueueing situation in real life. 8ne encounters a $ueue with a single server in many situations, including business %e.g. sales cler+(, industry %e.g. a production line(, transport %e.g. a bus, a ta-i ran+, an intersection(, telecommunications %e.g. &elephone line(, computing %e.g. processor sharing(. 'ven where there are multiple servers handling the situation it is possible to consider each server individually as part of the larger system, in many cases. %e.g supermar+et chec+out has several single server $ueues that the customer can select from.( *onse$uently, being able to model and analyse a single server $ueues behaviour is a particularly useful thing to do.
Poisson arri%als and ser%ice M+M++5+5 represents a single server that has unlimited $ueue capacity and infinite calling population, both arrivals and service are oisson %or random( processes, meaning the statistical distribution of both the inter)arrival times and the service times follow the e-ponential distribution. 9ecause of the mathematical nature of the e-ponential distribution, a number of $uite simple relationships are able to be derived for several performance measures based on +nowing the arrival rate and service rate.
21
&his is fortunate because, an ;:;: $ueuing model can be used to appro-imate many $ueuing situations.
Poisson arri%als and general ser%ice M+6++5+5 represents a single server that has unlimited $ueue capacity and infinite calling population, while the arrival is still oisson process, meaning the statistical distribution of the inter)arrival times still follow the e-ponential distribution, the distribution of the service time does not. &he distribution of the service time may follow any general statistical distribution, not Must e-ponential. Relationships are still able to be derived for a %limited( number of performance measures if one +nows the arrival rate and the mean and variance of the service rate.
number of special cases of ;:>: provide specific solutions that give broad insights into the best model to choose for specific $ueueing situations because they permit the comparison of those solutions to the performance of an ;:;: model.
M"lti$le-servers #"e"e ;ultiple %identical()servers $ueue situations are fre$uently encountered in telecommunications or a customer service environment. When modelling these situations care is needed to ensure that it is a multiple servers $ueue, not a networ+ of single server $ueues, because results may differ depending on how the $ueuing model behaves. 8ne observational insight provided by comparing $ueuing models is that a single $ueue with multiple servers performs better than each server having their own $ueue and that a single large pool of servers performs better than two or more smaller pools, even though there are the same total number of servers in the system. 8ne simple e-ample to prove the above fact is as follows" *onsider a system having ? input lines, single $ueue and ? servers.&he output line has a capacity of =1 +bit:s. *onsidering the arrival rate at each input as 2 pac+ets:s. 0o, the total arrival rate is = pac+ets:s. With an average of 2/// bits per pac+et, the service rate is =1 +bit:s:2///b 6 2 pac+ets:s.
%nfinitely many servers While never e-actly encountered in reality, an infinite-servers %e.g. M+M+5( model is a convenient theoretical model for situations that involve storage or delay, such as par+ing lots, warehouses and even atomic transitions. n these
25
models there is no $ueue, as such, instead each arriving customer receives service. When viewed from the outside, the model appears to delay or store each customer for some time. ueuein! System "lassification
With 7ittles &heorem, we have developed some basic understanding of a $ueueing system. &o further our understanding we will have to dig deeper into characteristics of a $ueueing system that impact its performance. For e-ample, $ueueing re$uirements of a restaurant will depend u pon factors li+e" •
•
•
$er%ice Process
•
&he probability density distribution that determines the customer arrivals in the system.
•
n a messaging system, this refers to the message arrival probability distribution. &he probability density distribution that determines the customer service times in the system.
•
•
Num7er $er%ers
o!
•
•
n a messaging system, this refers to the message transmission time distribution. 0ince message transmission is directly proportional to the length of the message, this parameter indirectly refers to the message length distribution. !umber of servers available to service the customers. n a messaging system, this refers to the number of lin+s between the source and destination nodes.
9ased on the above characteristics, $ueueing systems can be classified by the following convention" A+$+n
Where is the arrival process, 0 is the service process and n is the number of servers. and 0 are can be any of the following"
2=
; %;ar+ov( D %Deterministic( > %>eneral(
'-ponential probability density ll customers have the same value ny arbitrary probability distribution
'-amples of $ueueing systems that can be defined with this convention are" •
•
•
M+M+. &his is the simplest $ueueing system to analy4e.
;ar+ovian arrival processes n $ueuing theory, Marko%ian arri%al processes are used to model the arrival customers to $ueue. 0ome of the most common include the Poisson process, Mar$ovian arrival process and the *atch Mar$ovian arrival process. ;ar+ovian arrival processes has two processes. continuous)time ;ar+ov process +%t ( , a ;ar+ov process which is generated by a generator or rate matrix, ,. &he other process is a counting process %t (, which has state space %where is the set of all natural numbers(. %t ( increases every time there is a transition in +%t ( which mar+ed.
Poisson #rocess &he oisson arrival process or oisson process counts the number of arrivals, each of which has a e-ponentially distributed time between arrival. n the most general case this can be represented by the rate matri-,
Mar$ov arrival #rocess &he Marko% arri%al process % MAP ( is a generalisation of the oisson process by having non)e-ponential distribution soMourn between arrivals. &he homogeneous case has rate matri-,
2B
7ittles law n $ueueing theory, Little8s result , theorem, lemma, or law says" &he average number of customers in a stable system %over some time interval(, !, is e$ual to their average arrival rate, O, multiplied by their average time in the system, &, or"
lthough it loo+s intuitively reasonable, its a $uite remar+able result, as it implies that this behavior is entirely independent of any of the detailed probability distributions involved, and hence re$uires no assumptions about the schedule according to which customers arrive or are serviced, or whether they are served in the order in which they arrive. t is also a comparatively recent result ) it was first proved by ohn 7ittle, an nstitute rofessor and the *hair of ;anagement 0cience at the ;& 0loan 0chool of ;anagement, in =.
7et S%t ( be to some system in the interval T/, t U. 7et V%t ( be the number of departures from the same system in the interval T/, t U. 9oth S%t ( and V%t ( are integer valued increasing functions by their definition. 7et T t be the mean time spent in the system %during the interval T/, t U( for all the customers who were in the system during the interval T/, t U. 7et t be the mean number of customers in the system over the duration of the interval T/, t U. f the following limits e-ist,
and, further, if O 6 then 7ittles theorem holds, the limit
2?
e-ists and is given by 7ittles theorem,
Ideal Per!ormance
2
(ffects of "on!estion )
"on!estion*"ontrol Mechanisms
–
Backpressure Request from destination to source to reduce rate
–
Useful only on a logical connection basis
–
Requires hop-by-hop flow control mechanism
–
Policing Measuring and restricting packets as they enter the network hoke packet
–
!pecific message back to source
–
"#g#$ %MP !ource &uench
–
%mplicit congestion signaling !ource detects congestion from transmission delays and lost packets and reduces flow
/
(+#licit con!estion si!nalin!
Frame Relay reduces networ+ overhead by implementing simple congestion) notification mechanisms rather than e-plicit, per)virtual)circuit flow control. Frame Relay typically is implemented on reliable networ+ media, so data integrity is not sacrificed because flow control can be left to higher)layer protocols. Frame Relay implements two congestion)notification mechanisms" #
Forward)e-plicit congestion notification %F'*!(
#
9ac+ward)e-plicit congestion notification %9'*!(
F'*! and 9'*! each is controlled by a single bit contained in the Frame Relay frame header. &he Frame Relay frame header also contains a Discard 'ligibility %D'( bit, which is used to identify less important traffic that can be dropped during periods of congestion. &he ./" *it is part of the ddress field in the Frame Relay frame header. &he F'*! mechanism is initiated when a D&' device sends Frame Relay frames into the networ+. f the networ+ is congested, D*' devices %switches( set the value of the frames F'*! bit to . When the frames reach the destination D&' device, the ddress field %with the F'*! bit set( indicates that the frame e-perienced congestion in the path from source to destination. &he D&' device can relay this information to a higher)layer protocol for processing. Depending on the implementation, flow control may be initiated, or the indication may be ignored. &he 0/" *it is part of the ddress field in the Frame Relay frame header. D*' devices set the value of the 9'*! bit to in fra mes traveling in the opposite direction of frames with their F'*! bit set. &his informs the receiving D&' device that a particular path through the networ+ is congested. &he D&' device then can relay this information to a higher)layer protocol for processing. Depending on the implementation, flow)control may be initiated, or the indication may be ignored.
Frame Relay Discard &ligi'ility &he Discard /ligi*ility (D/) *it is used to indicate that a frame has lower importance than other frames. &he D' bit is part of the ddress field in the Frame Relay frame header. D&' devices can set the value of the D' bit of a frame to to indicate that the frame has lower importance than other frames. When the networ+ becomes congested, D*' devices will discard frames with the D' bit set before discarding those that do not. &his reduces the li+elihood of critical data being dropped by Frame Relay D*' devices during periods of congestion.
Frame Relay &rror Chec!ing Frame Relay uses a common error)chec+ing mechanism +nown as the cyclic redundancy chec$ (""). &he *R* compares two calculated values to determine whether errors occurred during the transmission from source to destination. Frame Relay reduces networ+ overhead by implementing error chec+ing rather than error correction. Frame Relay typically is implemented on reliable networ+ media, so data integrity is not sacrificed because error correction can be left to higher)layer protocols running on top of Frame Relay.
Tra!!ic Management "onsiderations
– –
– – – –
– –
in
"ongested
Network
9
$ome
Fairness Aarious flows should sufferY e$ually 7ast)in)first)discarded may not be fair Juality of 0ervice %Jo0( Flows treated differently, based on need Aoice, video" delay sensitive, loss insensitive File transfer, mail" delay insensitive, loss sensitive nteractive computing" delay and loss sensitive Reservations olicing" e-cess traffic discarded or handled on best)effort basis
Frame /elay "ongestion "ontrol
;inimi4e frame discard ;aintain Jo0 %per)connection bandwidth( ;inimi4e monopoli4ation of networ+ 0imple to implement, little overhead ;inimal additional networ+ traffic Resources distributed fairly 7imit spread of congestion
2
8perate effectively regardless of flow
"ongestion A%oidance with 2:plicit $ignaling
&wo general strategies considered"
–
–
– – –
Network /esponse
each frame handler monitors its $ueuing behavior and ta+es action use F'*!:9'*! bits some:all connections notified of congestion User (end1system) /esponse receipt of 9'*!:F'*! bits in frame
– – –
9'*! at sender" reduce transmission rate F'*! at receiver" notify peer %via 7F or higher layer( to restrict flow Frame /elay Tra!!ic /ate Management Parameters
–
*ommitted nformation Rate %*R( verage data rate in bits:second that the networ+ agrees to support for a connection Data Rate of 3ser ccess *hannel %ccess Rate(
–
–
–
Fi-ed rate lin+ between user and networ+ %for networ+ access( *ommitted 9urst 0i4e %9c( ;a-imum data over an interval agreed to by networ+ '-cess 9urst 0i4e %9e( ;a-imum data, above 9c, over an interval that networ+ will attempt to transfer
/elationship o! "ongestion Parameters
1
5
Unit III
T"P Flow "ontrol
3ses a form of sliding window Differs from mechanism used in 77*,
T"P 0eader Fields !or Flow "ontrol 0e$uence number %0!( of first octet in data segment c+nowledgement number %!( Window %W( c+nowledgement contains ! 6 i, W 6 M" 8ctets through 0! 6 i ) ac+nowledged ermission is granted to send W 6 M more octets, i.e., octets i through i C M ) T"P "redit Allocation Mechanism
"redit Allocation is Fle:i7le
=
0uppose last message 9 issued was ! 6 i, W 6 M
&o increase credit to + %+ I M( when no new data, 9 issues ! 6 i, W 6 + &o ac+nowledge segment containing m octets %m H M(, 9 issues ! 6 i C m, W 6 M
E m Flow "ontrol Perspecti%es
"redit Policy Receiver needs a policy for how much credit to give sender *onservative approach" grant credit up to limit of available buffer space ;ay limit throughput in long)delay situations
B
8ptimistic approach" grant credit based on e-pectation of freeing space before data arrives 2!!ect o! 3indow $i;e W 6 &* window si4e %octets( R 6 Data rate %bps( at &* source D 6 ropagation delay %seconds( fter &* source begins transmitting, it ta+es D seconds for first octet to arrive, and D seconds for ac+nowledgement to return &* source could transmit at most 2RD bits, or RD:1 octets
Normali;ed Throughput $
W I RD : 1
1W:RD
W H RD : 1
0 6
3indow $cale Parameter
"omplicating Factors ;ultiple &* connections are multiple-ed over same networ+ interface, reducing R and efficiency For multi)hop connections, D is the sum of delays across each networ+ plus delays at each router f source data rate R e-ceeds data rate on one of the hops, that hop will be a bottlenec+ 7ost segments are retransmitted, reducing throughput. mpact depends on retransmission policy /etransmission $trategy &* relies e-clusively on positive ac+nowledgements and retransmission on ac+nowledgement timeout
?
&here is no e-plicit negative ac+nowledgement Retransmission re$uired when" 0egment arrives damaged, as indicated by chec+sum error, causing receiver to discard segment 0egment fails to arrive
Timers timer is associated with each segment as it is sent f timer e-pires before segment ac+nowledged, sender must retransmit Key Design ssue" value of retransmission timer &oo small" many unnecessary retransmissions, wasting networ+ bandwidth &oo large" delay in handling lost segment Two $trategies &imer should be longer than round)trip delay %send segment, receive ac+( Delay is variable $trategies. Fi-ed timer daptive
Pro7lems with Adapti%e $cheme eer &* entity may accumulate ac+nowledgements and not ac+n owledge immediately For retransmitted segments, canZt tell whether ac+nowledgement is response to original transmission or retransmission !etwor+ conditions may change suddenly
Adapti%e /etransmission Timer verage Round)&rip &ime %R&&( KC R&&%K C ( 6 [ R&&%i( KC i6
6 KC
K R&%K( C KC
R&&%K C (
/F" <=* 2:ponential A%eraging 0moothed Round)&rip &ime %0R&&(
0R&&%K C ( 6 S \ 0R&&%K( C % E S( \ 0R&&%K C ( &he older the observation, the less it is counted in the average.
/F" <=* /etransmission Timeout R&8%K C ( 6 ;in%39, ;a-%79, V \ 0R&&%K C (((
39, 79" prechosen fi-ed upper and lower bounds '-ample values for S, V" /.? H S H /.
. H V H 2./
mplementation olicy 8ptions 0end Deliver ccept n)order n)window Retransmit First)only 9atch individual c+nowledge immediate cumulative
T"P "ongestion "ontrol Dynamic routing can alleviate congestion by spreading load more evenly 9ut only effective for unbalanced loads and brief surges in traffic *ongestion can only be controlled by limiting total amount of data entering networ+ *; source Juench message is crude and not effective R0A may help but not widely implemented
T"P "ongestion "ontrol is i!!icult is connectionless and stateless, with no provision for detecting or controlling congestion &* only provides end)to)end flow control !o cooperative, distributed algorithm to bind together various &* entities
T"P Flow and "ongestion "ontrol &he rate at which a &* entity can transmit is determined by rate of incoming *Ks to previous segments with new credit Rate of c+ arrival determined by round)trip path between source and destination 9ottlenec+ may be destination or internet 0ender cannot tell which 8nly the internet bottlenec+ can be due to congestion
1/
T"P $egment Pacing
T"P Flow and "ongestion "ontrol
/etransmission Timer Management &hree &echni$ues to calculate retransmission timer %R&8(" R&& Aariance 'stimation '-ponential R&8 9ac+off KarnZs lgorithm
1
/TT #ariance 2stimation (>aco7son?s Algorithm) sources of high variance in R&& f data rate relative low, then transmission delay will be relatively large, with larger variance due to variance in pac+et si4e 7oad may change abruptly due to other sources eer may not ac+nowledge segments immediately
>aco7son?s Algorithm 0R&&%K C ( 6 % E g( \ 0R&&%K( C g \ R&&%K C (
0'RR%K C ( 6 R&&%K C ( E 0R&&%K( 0D'A%K C ( 6 % E h( \ 0D'A%K( C h \]0'RR%K C (] R&8%K C ( 6 0R&&%K C ( C f \ 0D'A%K C ( g 6 /.25 h 6 /.25 f 6 2 or f 6 1 %most current implementations use f 6 1(
Two @ther Factors acobsonZs algorithm can significantly improve &* performance, but"
What R&8 to use for retransmitted segmentsQ !0W'R" e-ponential R&8 bac+off algorithm Which round)trip samples to use as input to acobsonZs algorithmQ !0W'R" KarnZs algorithm
2:ponential /T@ 'acko!! ncrease R&8 each time the same segment retransmitted E bac+off process ;ultiply R&8 by constant" R&8 6 $ \ R&8 $ 6 2 is called binary e-ponential bac+off 3hich /ound1trip $amples f an ac+ is received for retransmitted segment, there are 2 possibilities" c+ is for first transmission c+ is for second transmission &* source cannot distinguish 2 cases !o valid way to calculate R&&"
E E
From first transmission to ac+, or From second transmission to ac+Q
12
E
Barn?s Algorithm Do not use measured R&& to update 0R&& and 0D'A *alculate bac+off R&8 when a retransmission occurs 3se bac+off R&8 for segments until an ac+ arrives for a segment that has not been retransmitted &hen use acobsonZs algorithm to calculate R&8
3indow Management 0low start Dynamic window si4ing on congestion Fast retransmit Fast recovery 7imited transmit $low $tart awnd 6 ;!T credit, cwndU where awnd 6 allowed window in segments cwnd 6 congestion window in segments credit 6 amount of unused credit granted in most recent ac+ cwnd 6 for a new connection and increased by for each ac+ received, up to a ma-imum 2!!ect o! $low $tart
1
ynamic 3indow $i;ing on "ongestion lost segment indicates congestion rudent to reset cwsd 6 and begin slow start process ;ay not be conservative enough" easy to drive a networ+ into saturation but hard for the net to recoverY %acobson( nstead, use slow start with linear growth in cwnd Illustration o! $low $tart and "ongestion A%oidance
Fast /etransmit R&8 is generally noticeably longer than actual R&& f a segment is lost, &* may be slow to retransmit &* rule" if a segment is received out of order, an ac+ must be issued immediately for the last in)order segment Fast Retransmit rule" if 1 ac+s received for same segment, h ighly li+ely it was lost, so retransmit immediately, rather than waiting for timeout Fast /eco%ery When &* retransmits a segment using Fast Retransmit, a segment was assumed lost *ongestion avoidance measures are appropriate at this point '.g., slow)start:congestion avoidance procedure &his may be unnecessarily conservative since multiple ac+s indicate segments are getting through Fast Recovery" retransmit lost segment, cut cwnd in half, proceed with linear increase of cwnd &his avoids initial e-ponential slow)start
11
Limited Transmit f congestion window at sender is small, fast retransmit may not get triggered, e.g., cwnd 6 3nder what circumstances does sender have small congestion windowQ
s the problem commonQ
f the problem is common, why not reduce number of duplicate ac+s needed to trigger retransmitQ Limited Transmit Algorithm 0ender can transmit new segment when conditions are met" &wo consecutive duplicate ac+s are received Destination advertised window allows transmission of segment mount of outstanding data after sending is less than or e$ual to cwnd C 2 Per!ormance o! T"P o%er ATM
T"P+IP o%er AAL-+ATM
Per!ormance o! T"P o%er U'/ 9uffer capacity at &; switches is a critical parameter in assessing &* throughput performance nsufficient buffer capacity results in lost &* segments and retransmissions 2!!ect o! $witch 'u!!er $i;e Data rate of 1 ;bps 'nd)to)end propagation delay of = Ns
15
pac+et si4es of 52 octets to ?/ &* window si4es from ? Kbytes to =1 Kbytes &; switch buffer si4e per port from 25= cells to ?/// 8ne)to)one mapping of &* connections to &; virtual circuits &* sources have infinite supply of data ready @7ser%ations f a single cell is dropped, other cells in the same datagram are unusable, yet &; networ+ forwards these useless cells to destination 0maller buffer increase probability of dropped cells 7arger segment si4e increases number of useless cells transmitted if a single cell dropped Partial Packet and 2arly Packet iscard Reduce the transmission of useless cells Wor+ on a per)virtual circuit basis artial ac+et Discard
E
f a cell is dropped, then drop all subse$uent cells in that segment %i.e., loo+ for cell with 0D3 type bit set to one( 'arly ac+et Discard
E
When a switch buffer reaches a threshold level, preemptively discard all cells in a segment $electi%e rop deally, !:A cells buffered for each of the A virtual circuits W%i( 6 !%i( 6 !%i( \ A !:A ! f ! I R and W%i( I _ then drop ne-t new pac+et on A* i _ is a parameter to be chosen ATM $witch 'u!!er Layout
Fair 'u!!er Allocation ;ore aggressive dropping of pac+ets as congestion increases Drop new pac+et when"
1=
! I R and W%i( I _ \ 9 E R ! ) R T"P o%er A'/ >ood performance of &* over 39R can be achieved with minor adMustments to switch mechanisms &his reduces the incentive to use the more comple- and more e-pensive 9R service erformance and fairness of 9R $uite sensitive to some 9R parameter settings 8verall, 9R does not provide significant performance over simpler and less e-pensive 39R)'D or 39R)'D)F9
Tra!!ic and "ongestion "ontrol in ATM Networks Introduction *ontrol needed to prevent switch buffer overflow uaranteed Frame Rate %>FR( /e4uirements !or ATM Tra!!ic and "ongestion "ontrol ;ost pac+et switched and frame relay networ+s carry non)real)time bursty data E !o need to replicate timing at e-it node E 0imple statistical multiple-ing E 3ser !etwor+ nterface capacity slightly greater than average of channels *ongestion control tools from these technologies do not wor+ in &; Pro7lems with ATM "ongestion "ontrol ;ost traffic not amenable to flow control E Aoice ` video can not stop generating Feedbac+ slow E 0mall cell transmission time v propagation delay Wide range of applications E From few +bps to hundreds of ;bps E Different traffic patterns
1B
E Different networ+ services
Network "ontri7ution to "ell elay #ariation n pac+et switched networ+ E Jueuing effects at each intermediate switch E rocessing time for header and routing 7ess for &; networ+s E ;inimal processing overhead at switches Fi-ed cell si4e, header format
1?
!o flow control or error control processing E &; switches have e-tremely high throughput E *ongestion can cause cell delay variation 9uild up of $ueuing effects at switches &otal load accepted by networ+ must be controlled "ell elay #ariation at UNI *aused by processing in three layers of &; model E 0ee ne-t slide for details !one of these delays can be predicted !one follow repetitive pattern 0o, random element e-ists in time interval between reception by &; stac+ and transmission ATM Tra!!ic1/elated Attri7utes 0i- service categories %see chapter 5( E *onstant bit rate %*9R( E Real time variable bit rate %rt)A9R( E !on)real)time variable bit rate %nrt)A9R( E 3nspecified bit rate %39R( E vailable bit rate %9R( E >uaranteed frame rate %>FR( *haracteri4ed by &; attributes in four categories E &raffic descriptors E Jo0 parameters E *ongestion E 8ther Tra!!ic Parameters
&raffic pattern of flow of cells E ntrinsic nature of traffic 0ource traffic descriptor E ;odified inside networ+ *onnection traffic descriptor $ource Tra!!ic escriptor ea+ cell rate E 3pper bound on traffic that can be submitted E Defined in terms of minimum spacing between cells T E *R 6 :T E ;andatory for *9R and A9R services 0ustainable cell rate E 3pper bound on average rate E *alculated over large time scale relative to T E Re$uired for A9R E 'nables efficient allocation of networ+ resources between A9R sources E 8nly useful if 0*R H *R ;a-imum burst si4e E ;a- number of cells that can be sent at *R E f bursts are at ;90, idle gaps must be enough to +eep overall rate below 0*R
1
E Re$uired for A9R ;inimum cell rate E ;in commitment re$uested of networ+ E *an be 4ero E 3sed with 9R and >FR E 9R ` >FR provide rapid access to spare networ+ capacity up to *R E *R E ;*R represents elastic component of data flow E 0hared among 9R and >FR flows ;a-imum frame si4e E ;a- number of cells in frame that can be carried over >FR connection E 8nly relevant in >FR "onnection Tra!!ic escriptor ncludes source traffic descriptor plus") *ell delay variation tolerance mount of variation in cell delay introduced by networ+ interface and 3! 9ound on delay variability due to slotted nature of &;, physical layer overhead and layer functions %e.g. cell multiple-ing( Represented by time variable *onformance definition 0pecify conforming cells of connection at 3! 'nforced by dropping or mar+ing cells over definition &uality o! $er%ice Parameters1ma:"T *ell transfer delay %*&D( &ime between transmission of first bit of cell at source and reception of last bit at destination &ypically has probability density function %see ne-t slide( Fi-ed delay due to propagation etc. *ell delay variation due to buffering and scheduling ;a-imum cell transfer delay %ma-*&D(is ma- re$uested delay for connection Fraction S of cells e-ceed threshold Discarded or delivered late Peak1to1peak "# C "L/ ea+)to)pea+ *ell Delay Aariation Remaining %)S( cells within Jo0 Delay e-perienced by these cells is between fi-ed delay and ma-*&D &his is pea+)to)pea+ *DA *DA& is an upper bound on *DA *ell loss ratio Ratio of cells lost to cells transmitted
"ell Trans!er elay PF
5/
"ongestion "ontrol Attri7utes 8nly feedbac+ is defined 9R and >FR ctions ta+en by networ+ and end systems to regulate traffic submitted 9R flow control daptively share available bandwidth @ther Attri7utes 9ehaviour class selector %9*0( E 0upport for differentiated services %chapter =( E rovides different service levels among 39R connections E ssociate each connection with a behaviour class E ;ay include $ueuing and scheduling ;inimum desired cell rate Tra!!ic Management Framework 8bMectives of &; layer traffic and congestion control E 0upport Jo0 for all foreseeable services E !ot rely on networ+ specific 7 protocols nor higher layer application specific protocols E ;inimi4e networ+ and end system comple-ity E ;a-imi4e networ+ utili4ation Timing Le%els *ell insertion time Round trip propagation time *onnection duration 7ong term
Tra!!ic "ontrol and "ongestion Functions
5
Tra!!ic "ontrol $trategy Determine whether new &; connection can be accommodated gree performance parameters with subscriber &raffic contract between subscriber and networ+ &his is congestion avoidance f it fails congestion may occur E nvo+e congestion control Tra!!ic "ontrol Resource management using virtual paths *onnection admission control 3sage parameter control 0elective cell discard &raffic shaping '-plicit forward congestion indication /esource Management Using #irtual Paths llocate resources so that traffic is separated according to service characteristics Airtual path connection %A*( are groupings of virtual channel connections %A**( Applications 3ser)to)user applications E A* between 3! pair E !o +nowledge of Jo0 for individual A** E 3ser chec+s that A* can ta+e A**sZ demands 3ser)to)networ+ applications E A* between 3! and networ+ node E !etwor+ aware of and accommodates Jo0 of A**s !etwor+)to)networ+ applications E A* between two networ+ nodes E !etwor+ aware of and accommodates Jo0 of A**s
52
/esource Management "oncerns *ell loss ratio ;a- cell transfer delay ea+ to pea+ cell delay variation ll affected by resources devoted to A* f A** goes through multiple A*s, performance depends on consecutive A*s and on node performance E A* performance depends on capacity of A* and traffic characteristics of A**s E A** related function depends on switching:processing speed and priority #""s and #P"s "on!iguration
Allocation o! "apacity to #P" ggregate pea+ demand E ;ay set A* capacity %data rate( to total of A** pea+ rates 'ach A** can give Jo0 to accommodate pea+ demand A* capacity may not be fully used 0tatistical multiple-ing E A* capacity I6 average data rate of A**s but H aggregate pea+ demand E >reater *DA and *&D E ;ay have greater *7R E ;ore efficient use of capacity E For A**s re$uiring lower Jo0 E >roup A**s of similar traffic together
"onnection Admission "ontrol 3ser must specify service re$uired in both directions E *ategory
5
E
*onnection traffic descriptor 0ource traffic descriptor *DA& Re$uested conformance definition E Jo0 parameter re$uested and acceptable value !etwor+ accepts connection only if it can commit resources to support re$uests Procedures to $et Tra!!ic "ontrol Parameters
"ell Loss Priority &wo levels re$uested by user E riority for individual cell indicated by *7 bit in header E f two levels are used, traffic parameters for both flows specified
Location o! UP" Function
51
Peak "ell /ate Algorithm eneric cell rate algorithm E 7ea+y buc+et algorithm E 6eneric "ell /ate Algorithm
#irtual $cheduling Algorithm
55
Leaky 'ucket Algorithm
"ontinuous Leaky 'ucket Algorithm
$ustaina7le "ell /ate Algorithm 8perational definition of relationship between sustainable cell rate and burst tolerance 3sed by 3* to monitor compliance 0ame algorithm as pea+ cell rate
UP" Actions *ompliant cell pass, non)compliant cells discarded
5=
f no additional resources allocated to *76 traffic, *76/ cells * f two level cell loss priority cell with" E *76/ and conforms passes E *76/ non)compliant for *76/ traffic but compliant for *76/C is tagged and passes E *76/ non)compliant for *76/ and *76/C traffic discarded E *76 compliant for *76/C passes E *76 non)compliant for *76/C discarded Possi7le Actions o! UP"
2:plicit Forward "ongestion Indication 'ssentially same as frame relay f node e-periencing congestion, set forward congestion indication is cell headers E &ells users that congestion avoidance should be initiated in this direction E 3ser may ta+e action at higher level A'/ Tra!!ic Management Jo0 for *9R, A9R based on traffic contract and 3* described previously !o congestion feedbac+ to source 8pen)loop control !ot suited to non)real)time applications E File transfer, web access, R*, distributed file systems E !o well defined traffic characteristics e-cept *R E *R not enough to allocate resources 3se best efforts or closed)loop control 'est 2!!orts 0hare unused capacity between applications s congestion goes up" E *ells are lost E 0ources bac+ off and reduce rate E Fits well with &* techni$ues %chapter 2(
5B
E
nefficient *ells dropped causing re)transmission "losed1Loop "ontrol 0ources share capacity not used by *9R and A9R rovide feedbac+ to sources to adMust load void cell loss 0hare capacity fairly 3sed for 9R "haracteristics o! A'/ 9R connections share available capacity E ccess instantaneous capacity unused by *9R:A9R E ncreases utili4ation without affecting *9R:A9R Jo0 0hare used by single 9R connection is dynamic E Aaries between agreed ;*R and *R !etwor+ gives feedbac+ to 9R sources E 9R flow limited to available capacity E 9uffers absorb e-cess traffic prior to arrival of feedbac+ 7ow cell loss E ;aMor distinction from 39R Feed7ack Mechanisms *ell transmission rate characteri4ed by" E llowable cell rate *urrent rate E ;inimum cell rate ;in for *R ;ay be 4ero E ea+ cell rate ;a- for *R E nitial cell rate 0tart with *R6*R dMust *R based on feedbac+ Feedbac+ in resource management %R;( cells E *ell contains three fields for feedbac+ *ongestion indicator bit %*( !o increase bit %!( '-plicit cell rate field %'R( $ource /eaction to Feed7ack f *6 E Reduce *R by amount proportional to current *R but not less than *R 'lse if !6/ E ncrease *R by amount proportional to *R but not more than *R f *RI'R set *RH)ma-T'R,;*RU "ell Flow on A'/ &wo types of cell E Data ` resource management %R;( 0ource receives regular R; cells
5?
E Feedbac+ 9ul+ of R; cells initiated by source E 8ne forward R; cell %FR;( per %!rm)( data cells !rm preset E usually 2 E 'ach FR; is returned by destination as bac+wards R; %9R;( cell E FR; typically *6/, !6/ or 'R desired transmission rate in range *RH6'RH6*R E ny field may be changed by switch or destination before return ATM $witch /ate "ontrol Feed7ack 'F* mar+ing '-plicit forward congestion indication *auses destination to set * bit in 'R; Relative rate mar+ing 0witch directly sets * or ! bit of R; f set in FR;, remains set in 9R; Faster response by setting bit in passing 9R; Fastest by generating new 9R; with bit set '-plicit rate mar+ing 0witch reduces value of 'R in FR; or 9R; Flow o! ata and /M "ells
A/' Feed7ack % T"P A"B 9R feedbac+ controls rate of transmission E Rate control &* feedbac+ controls window si4e E *redit control R9 feedbac+ from switches or destination &* feedbac+ from destination only /M "ell Format
5
/M "ell Format Notes &; header has &6/ to indicate R; cell 8n virtual channel A and A* same as data cells on connection 8n virtual path A same, A*6= rotocol id identifies service using R; %R96( ;essage type E Direction FR;6/, 9R;6 E 9'*! cell. 0ource %9!6/( or switch:destination %9!6( E * %6 for congestion( E ! %6 for no increase( E Re$uest:c+nowledge %not used in &; forum spec(
A/' Parameters
=/
A/' "apacity Allocation &; switch must perform" *ongestion control ;onitor $ueue length Fair capacity allocation &hrottle bac+ connections using more than fair share &; rate control signals are e-plicit &* are implicit ncreasing delay and cell loss "ongestion "ontrol Algorithms1'inary Feed7ack 3se only 'F*, * and ! bits 0witch monitors buffer utili4ation When congestion approaches, binary notification E 0et 'F* on forward data cells or * or ! on FR; or 9R; &hree approaches to which to notify E 0ingle FF8 $ueue E ;ultiple $ueues E Fair share notification $ingle FIF@ &ueue
=
When buffer use e-ceeds threshold %e.g. ?/( E 0witch starts issuing binary notifications E *ontinues until buffer use falls below threshold E *an have two thresholds 8ne for start and one for stop 0tops continuous on:off switching E 9iased against connections passing through more switches Multiple &ueues 0eparate $ueue for each A* or group of A*s 0eparate threshold on each $ueue 8nly connections with long $ueues get binary notifications E Fair E 9adly behaved source does not affect other A*s E Delay and loss behaviour of individual A*s separated *an have different Jo0 on different A*s Fair $hare
0elective feedbac+ or intelligent mar+ing &ry to allocate capacity dynamically '.g. fairshare 6%target rate(:%number of connections( ;ar+ any cells where **RIfairshare 2:plicit /ate Feed7ack $chemes *ompute fair share of capacity for each A* Determine current load or congestion *ompute e-plicit rate %'R( for each connection and send to source &hree algorithms E 'nhanced proportional rate control algorithm 'R* E '-plicit rate indication for congestion avoidance 'R* E *ongestion avoidance using proportional control ** 2nhanced Proportional /ate "ontrol Algorithm(2P/"A 0witch trac+s average value of current load on each connection E ;ean allowed cell rate %;R*( E ;*R% (6%)S(%;*R% )( C S**R% ( E **R% ( is **R field in th FR; E &ypically S6:= E 9ias to past values of **R over current E >ives estimated average load passing through switch E f congestion, switch reduces each A* to no more than DF;*R DF6down pressure factor, typically B:? 'RH)minT'R, DF;*RU Load Factor dMustments based on load factor 7F6nput rate:target rate
=2
E E E
nput rate measured over fi-ed averaging interval &arget &arget rate slightly below lin+ bandwidth %?5 to /( 7FI congestion threatened A*s will have to reduce rate
2:plicit /ate Indication !or "ongestion A%oidance (2/I"A) ttempt to +eep 7F close to Define" fairshare 6 %target rate(:%number of connections( A*share 6 **R:7F 6 %**R:%nput Rate(( %&arget %&arget Rate( 'R* selectively adMusts A* rates E &otal &otal 'R allocated to connections matches target rate E llocation is fair E 'R 6 ma-Tfairshare, A*shareU E A*s whose A*share is less than than their fairshare get greater increase "ongestion A%oidance Using Proportional "ontrol ("AP") fairshareH)fairshareminT'R3,C%)7F(RupU nT'R3,C%)7F(RupU f 7FH fairshareH)fairsharemi fairshareH)fairshareminT'R3,)%)7F(RdnU nT'R3,)%)7F(RdnU f 7FI fairshareH)fairsharemi 'R3I, determines ma- increase Rup between /./25 and /., slope parameter Rdn, between /.2 and /.?, slope parameter 'RF typically /.5, ma- decrease in allottment of fair share f fairshare H 'R value in R; cells, 'RH)fairshare 0impler than 'R* *an show large rate oscillations if RF %Rate increase factor( too high *an lead to unfairness 6/F @%er%iew 0imple as 39R from end system view E 'nd system does no policing or traffic shaping E ;ay transmit at line rate of &; adaptor ;odest re$uirements on &; networ+ !o guarantee of frame delivery
=
Mechanisms !or supporting /ate 6uarantees &agging &agging and policing 9uffer management 0cheduling Tagging Tagging and an d Policing &agging &agging identifies frames that conform con form to contract and those that donZt E *76 for those that donZt 0et by networ+ element doing conformance chec+ ;ay be networ+ element or source showing less important frames E >et lower Jo0 in buffer management and scheduling E &agged &agged cells can be discarded at ingress to & &; networ+ or subse$uent switch E Discarding is a policing function 'u!!er Management &reatment of cells in buffers or when arriving and re$uiring buffering bu ffering f congested %high buffer occupancy( tagged cells discarded in preference to untagged Discard tagged cell to ma+e room for untagged cell ;ay buffer per)A* Discards may be based on per $ueue thresholds $cheduling >ive preferential treatment to untagged cells 0eparate $ueues for each A* E er A* scheduling decisions E '.g. FF8 modified to give *76/ cells higher priority 0cheduling between $ueues controls outgoing rate of A*s E ndividual cells get fair allocation while meeting traffic traffic contract
"omponents o! 6F/ Mechanism
=1
6F/ "on!ormance e!inition 3* function E 3* monitors A* for traffic traffic conformance E &ag &ag or discard non)conforming cells Frame conforms if all cells in frame conform E Rate of cells within contract >eneric cell rate algorithm *R and *DA& specified for connection E ll cells have same *7 E Within Within ma-imum frame si4e %;F0( &o$ 2ligi7ility 2ligi7ili ty Test Test &est &est for contract conformance E Discard or tag non)conforming cells 7oo+ing at upper bound on traffic E Determine frames eligible for Jo0 guarantee 3nder >FR contract for A* 7oo+ing at lower bound for traffic Frames are one of" E !onconforming" cells tagged or discarded E *onforming ineligible" best efforts efforts E *onforming eligible" guaranteed delivery
$impli!ied Frame 'ased 6"/A
=5
Unit I# Integrated and i!!erentiated $er%ices Introduction !ew additions to nternet increasing traffic E raphics Real time voice and video E !eed to manage traffic and control congestion 'F& standards E ntegrated services *ollective service to set of traffic demands in domain E 7imit demand ` reserve resources E Differentiated services *lassify traffic in groups Different group traffic handled differently
Integrated $er%ices Architecture (I$A)
E
v1 header fields for precedence and type of service usually ignored &; only networ+ designed to support &*, 3D and real)time traffic ;ay need new installation !eed to support Juality of 0ervice %Jo0( within &*:
E dd functionality to routers E ;eans of re$uesting Jo0 Internet Tra!!ic 9 2lastic
E E
E E
E E E
*an adMust to changes in delay and throughput '.g. common &* and 3D application ');ail E insensitive to delay changes F& E 3ser e-pect delay proportional to file si4e 0ensitive to changes in throughput 0!; E delay not a problem, e-cept when caused by congestion Web %<&&(, &'7!'& E sensitive to delay !ot per pac+et delay E total elapsed time '.g. web page loading time For small items, delay across internet dominates For large items it is throughput over connection !eed some Jo0 control to match to demand
==
Internet Tra!!ic 9 Inelastic Does not easily adapt to changes in delay and throughput E Real time traffic &hroughput E ;inimum may be re$uired Delay E '.g. stoc+ trading itter ) Delay variation E ;ore Mitter re$uires a bigger buffer E '.g. teleconferencing re$uires reasonable upper bound ac+et loss Inelastic Tra!!ic Pro7lems Difficult to meet re$uirements on networ+ with variable $ueuing delays and congestion !eed preferential treatment pplications need to state re$uirements E head of time %preferably( or on the fly E 3sing fields in header E Resource reservation protocol ;ust still support elastic traffic E Deny service re$uests that leave too few resources to handle elastic traffic demands I$A Approach rovision of Jo0 over 0haring available capacity when congested Router mechanisms
E
E
Routing lgorithms 0elect to minimi4e delay ac+et discard *auses &* sender to bac+ off and reduce load 'nahnced by 0
Flow
pac+et can be associated with a flow Distinguishable stream of related pac+ets E E From single user activity E Re$uiring same Jo0 '.g. one transport connection or one video stream E E 3nidirectional E *an be more than one recipient ;ulticast E ;embership of flow identified by source and destination address, port numbers, protocol type E v= header flow identifier can be used but isnot necessarily e$uivalent to 0 flow I$A Functions dmission control E For Jo0, reservation re$uired for new flow
=B
E
E
E
E E
R0A used Routing algorithm 9ase decision on Jo0 parameters Jueuing discipline &a+e account of different flow re$uirements Discard policy ;anage congestion ;eet Jo0
I$A Implementation in /outer 9ac+ground Functions
Forwarding functions
I$A "omponents 9 'ackground Functions Reservation rotocol
E
R0A dmission control ;anagement agent
E
*an use agent to modify traffic control database and direct admission control Routing protocol I$A "omponents 9 Forwarding *lassifier and route selection E ncoming pac+ets mapped to classes 0ingle flow or set of flows with same Jo0 E '.g. all video flows 9ased on header fields E Determines ne-t hop ac+et scheduler
=?
E E
E
;anages one or more $ueues for each output 8rder $ueued pac+ets sent 9ased on class, traffic control database, current and past activity on outgoing port olicing
I$A $er%ices &raffic specification %&0pec( defined as service for flow 8n two levels
E
>eneral categories of service >uaranteed *ontrolled load 9est effort %default( E articular flow within category &0pec is part of contract Token 'ucket ;any traffic sources can be defined by to+en buc+et scheme rovides concise description of load imposed by flow
E
'asy to determine resource re$uirements rovides input parameters to policing function Token 'ucket iagram
I$A $er%ices 9 6uaranteed $er%ice ssured capacity level or data rate 0pecific upper bound on $ueuing delay through networ+ E ;ust be added to propagation delay or latency to get total delay E 0et high to accommodate rare long $ueue delays !o $ueuing losses E .e. no buffer overflow '.g. Real time play bac+ of incoming signal can use delay buffer for incoming signal but will not tolerate pac+et loss
=
I$A $er%ices 9 "ontrolled Load &ightly appro-imates to best efforts under unloaded conditions !o upper bound on $ueuing delay E reedy &* connection can crowd out altruistic connections E f one connection does not bac+ off, others may bac+ off more Fair &ueuing (F&) ;ultiple $ueues for each port E 8ne for each source or flow E Jueues services round robin E 'ach busy $ueue %flow( gets e-actly one pac+et per cycle E 7oad balancing among flows E !o advantage to being greedy our $ueue gets longer, increasing your delay E 0hort pac+ets penali4ed as each $ueue sends one pac+et per cycle FIF@ and F&
Processor $haring ;ultiple $ueues as in FJ 0end one bit from each $ueue per round
E
7onger pac+ets no longer get an advantage *an wor+ out virtual %number of cycles( start and finish time for a given pac+et
B/
ood appro-imation to performance of 0
E
&hroughput and delay converge as time increases
"omparison o! FIF@D F& and '/F&
6enerali;ed Processor $haring (6P$) 9RFJ can not provide different capacities to different flows 'nhancement called Weighted fair $ueue %WFJ( From 0, allocate weighting to each flow that determines how many bots are sent during each round E f weighted 5, then 5 bits are sent per round >ives means of responding to different service re$uests >uarantees that delays do not e-ceed bounds 3eighted Fair &ueue 'mulates bit by bit >0 0ame strategy as 9RFJ
B
FIF@ % 3F&
E E E
Proacti%e Packet iscard *ongestion management by proactive pac+et discard
9efore buffer full 3sed on single FF8 $ueue or multiple $ueues for elastic traffic '.g. Random 'arly Detection %R'D(
/andom 2arly etection (/2) Moti%ation 0urges fill buffers and cause discards 8n &* this is a signal to enter slow start phase, reducing load E 7ost pac+ets need to be resent
B2
E
dds to load and delay >lobal synchroni4ation &raffic burst fills $ueues so pac+ets lost ;any &* connections enter slow start &raffic drops so networ+ under utili4ed *onnections leave slow start at same time causing burst 9igger buffers do not help &ry to anticipate onset of congestion and tell one connection to slow down
/2 esign 6oals
E
E
*ongestion avoidance >lobal synchroni4ation avoidance *urrent systems inform connections to bac+ off implicitly by dropping pac+ets voidance of bias to bursty traffic Discard arriving pac+ets will do this 9ound on average $ueue length
E
/2 Algorithm etail
B
B1
i!!erentiated $er%ices ($)
E E
tool E
0 and R0A comple- to deploy ;ay not scale well for large volumes of traffic mount of control signals ;aintenance of state information at routers D0 architecture designed to provide simple, easy to implement, low overhead 0upport range of networ+ services Differentiated on basis of performance
"haracteristics o! $ 3se v1 header &ype of 0ervice or v= &raffic *lass field E !o change to 0ervice level agreement %07( established between provider %internet domain( and customer prior to use of D0 E D0 mechanisms not needed in applications 9uild in aggregation E ll traffic with same D0 field treated same '.g. multiple voice connections D0 implemented in individual routers by $ueuing and forwarding based on E D0 field 0tate information on flows not saved by routers
$er%ices rovided within D0 domain E *ontiguous portion of nternet over which consistent set of D0 policies administered &ypically under control of one administrative entity E Defined in 07 *ustomer may be user organi4ation or other D0 d omain E E ac+et class mar+ed in D0 field 0ervice provider configures forwarding policies routers E 8ngoing measure of performance provided for each class D0 domain e-pected to provide agreed service internally f destination in another domain, D0 domain attempts to forward pac+ets through other domains E ppropriate service level re$uested from each domain $LA Parameters Detailed service performance parameters
E
E
&hroughput, drop probability, latency *onstraints on ingress and egress points ndicate scope of service
B5
&raffic profiles to be adhered to
E
&o+en buc+et Disposition of traffic in e-cess of profile 2:ample $er%ices Jualitative E " 7ow latency E 9" 7ow loss Juantitative E *" / in)profile traffic delivered with no more than 5/ms latency E D" 5 in)profile traffic delivered ;i-ed E '" &wice bandwidth of F E F" &raffic with drop precedence X has higher delivery probability than that with drop precedence D0 Field Detail 7eftmost = bits are D0 codepoint E =1 different classes available E pools -----/ " reserved for standards ////// " default pac+et class E E ---/// " reserved for bac+wards compatibility with v1 &80 ---- " reserved for e-perimental or local use ----/ " reserved for e-perimental or local use but may be allocated for future standards if needed Rightmost 2 bits unused
"on!iguration iagram
"on!iguration 9 Interior /outers Domain consists of set of contiguous routers nterpretation of D0 codepoints within domain is consistent
B=
nterior nodes %routers( have simple mechanisms to handle pac+ets based on codepoints E Jueuing gives preferential treatment depending on codepoint er
Per 0op 'eha%iour 9 2:pedited !orwarding remium service E 7ow loss, delay, MitterG assured bandwidth end)to)end service through domains E 7oo+s li+e point to point or leased line E Difficult to achieve E *onfigure nodes so traffic aggregate has well defined minimum departure rate 'F <9 E *ondition aggregate so arrival rate at any node is always less that minimum departure rate 9oundary conditioners Per 0op 'eha%iour 9 2:plicit Allocation 0uperior to best efforts Does not re$uire reservation of resources Does not re$uire detailed discrimination among flows 3sers offered choice of number of classes ;onitored at boundary node n or out depending on matching profile or not E nside networ+ all traffic treated as single pool of pac+ets, distinguished only as in or out
BB
Drop out pac+ets before in pac+ets if necessary Different levels of service because different number of in pac+ets for each user
P0' 1 Assured Forwarding Four classes defined
E
0elect one or more to meet re$uirements Within class, pac+ets mar+ed by customer or provider with one of three drop precedence values E 3sed to determine importance when dropping pac+ets as result of congestion
"odepoints !or AF P0'
B?
Unit # Protocols !or &o$ $upport Increased emands !eed to incorporate bursty and stream traffic in &*: architecture capacity ncrease E Faster lin+s, switches, routers E ntelligent routing policies E 'nd)to)end flow control ;ulticasting Juality of 0ervice %Jo0( capability &ransport protocol for streaming /esource /eser%ation 1 Unicast revention as well as reaction to congestion re$uired *an do this by resource reservation 3nicast E 'nd users agree on Jo0 for tas+ and re$uest from networ+ E ;ay reserve resources E Routers pre)allocate resources E f Jo0 not available, may wait or try at reduced Jo0 /esource /eser%ation 9 Multicast >enerate vast traffic E
B
Receivers can select one of multiple sources %channel selection( Deal gracefully with changes in routes E Re)establish reservations *ontrol protocol overheadndependent of routing protocol /$#P "haracteristics 3nicast and ;ulticast 0imple E 3nidirectional data flow E 0eparate reservations in two directions Receiver initiated E Receiver +nows which subset of source transmissions it wants ;aintain soft state in internet E Responsibility of end users roviding different reservation styles E 3sers specify how reservations for groups are aggregated &ransparent operation through non)R0A routers 0upport v1 %&o0 field( and v= %Flow label field( ata Flows 1 $ession Data flow identified by destination Resources allocated by router for duration of session Defined by E Destination address 3nicast or multicast E protocol identifier &*, 3D etc. E Destination port ;ay not be used in multicast Flow escriptor Reservation Re$uest E Flow spec Desired Jo0 3sed to set parameters in nodeZs pac+et scheduler 0ervice class, Rspec %reserve(, &spec %traffic( E Filter spec 0et of pac+ets for this reservation 0ource address, source prot Treatment o! Packets o! @ne $ession at @ne /outer
?/
/$#P @peration iagram
/$#P @peration >, >2, > members of multicast group 0, 02 sources transmitting to that group has reservation filter spec including 0 and 02 >, >2 from 0 only
?
R delivers from 02 to > but does not forward to R1 >, >2 send R0A re$uest with filter e-cluding 02 >, >2 only members of group reached through R1 E R1 doesnZt need to forward pac+ets from this session E R1 merges filter spec re$uests and sends to R R no longer forwards this sessionZs pac+ets to R1 E E 0tores filter spec but doesnZt propagate it
/eser%ation $tyles
Determines manner in which resource re$uirements from members of group are aggregated Reservation attribute E Reservation shared among senders %shared( *haracteri4ing entire flow received on multicast address E llocated to each sender %distinct( 0imultaneously capable of receiving data flow from each sender 0ender selection E 7ist of sources %e-plicit( E ll sources, no filter spec %wild card(
/eser%ation Attri7utes and $tyles
Reservation ttribute E Distinct 0ender selection e-plicit 6 Fi-ed filter %FF( 0ender selection wild card 6 none E 0hared 0ender selection e-plicit6 0hared)e-plicit %0'( 0ender selection wild card 6 Wild card filter %WF(
3ild "ard Filter $tyle
0ingle resource reservation shared by all senders to this address f used by all receivers" shared pipe whose capacity is largest of resource re$uests from receivers downstream from any point on tree ndependent of number of senders using it ropagated upstream to all senders WF%J( E 6 wild card sender E J 6 flowspec udio teleconferencing with multiple sites
Fi:ed Filter $tyle
Distinct reservation for each sender
?2
'-plicit list of senders FF%0J, 02J2,^( Aideo distribution
$hared 2:plicit $tyle
0ingle reservation shared among specific list of senders 0'%0, 02, 0, ^J( ;ulticast applications with multiple data sources but unli+ely to transmit simultaneously
/$#P Protocol Mechanisms
&wo message types E Resv 8riginate at multicast group receivers ropagate upstream ;erged and pac+et when appropriate *reate soft states Reach sender E llow host to set up traffic control for first hop E ath rovide upstream routing information ssued by sending hosts &ransmitted through distribution tree to all destinations
/$#P 0ost Model
Summary
R0A is a transport layer protocol that enables a networ+ to provide differentiated levels of service to specific flows of data. 8stensibly, different application types have different performance re$uirements. R0A ac+nowledges these differences and provides the mechanisms necessary to detect the levels of performance re$uired by different appli) cations and to modify networ+ behaviors to accommodate those re$uired levels. 8ver time, as time and latency)sensitive applications mature and proliferate, R0As capabilities will become increasingly important.
?
Revie, uestions
&E s it necessary to migrate a2ay from your existing routing protocol to support #3P4 AE R0A is not a routing protocol. nstead, it was designed to wor+ in conMunction with e-isting routing protocols. &hus, it is not necessary to migrate to a new routing protocol to support R0A. &E dentify the three #3P levels of service5 and explain the difference among them6 AE R0As three levels of service include best)effort, rate)sensitive, and delay)sensitive service. 9est)effort service is used for applications that re$uire reliable delivery rather than a timely delivery. Rate)sensitive service is used for any traffic that is sensitive to variation in the amount of bandwidth available. 0uch applications include <.2 videoconferencing, which was designed to run at a nearly constant rate. R0As third level of service is delay)sensitive service. Delay)sensitive traffic re$uires timely but not reliable delivery of data. &E 7hat are the t2o #3P reservation classes5 and ho2 do they differ4 AE reservation style is a set of control options that defines how a reservation operates. R0A supports two primary types of reservation styles" distinct reservations and shared reservations. distinct reservation establishes a flow for each sending device in a session. 0hared reservations aggregate communications flows for a set of senders. 'ach of these two reservation styles is defined by a series of filters. &E 7hat are #3P filters4 AE filter in R0A is a specific set of control options that specifies operational parameters for a reservation. R0As styles include wildcard)filter %WF(, fi-ed)filter %FF(, and shared)e-plicit %0'( filters.
8o2 can #3P *e used through net2or$ regions that do not support #3P4 &E AE R0A supports tunneling through networ+ regions that do not support R0A. &his capability was developed to enable a phased)in implementation of R0A.
Multiprotocol La7el $witching (MPL$)
Routing algorithms provide support for performance goals E Distributed and dynamic React to congestion 7oad balance across networ+ E 9ased on metrics Develop information that can be used in handling different service needs 'nhancements provide direct support E 0, D0, R0A
?1
!othing directly improves throughput or delay ;70 tries to match &; Jo0 support
'ackground
'fforts to marry and &; switching %psilon( &ag switching %*isco( ggregate route based switching %9;( *ascade % navigator( ll use standard routing protocols to define paths between end points ssign pac+ets to path as they enter networ+ 3se &; switches to move pac+ets along paths E &; switching %was( much faster than routers E 3se faster technology
e%elopments
'&F wor+ing group in B, proposed standard 2// Routers developed to be as fast as &; switches E Remove the need to provide both technologies in same networ+ ;70 does provide new capabilities E Jo0 support E &raffic engineering E Airtual private networ+s E ;ultiprotocol support
"onnection @riented &o$ $upport >uarantee fi-ed capacity for specific applications *ontrol latency:Mitter 'nsure capacity for voice rovide specific, guaranteed $uantifiable 07s *onfigure varying degrees of Jo0 for multiple customers ;70 imposes connection oriented framewor+ on based internets Tra!!ic 2ngineering bility to dynamically define routes, plan resource commitments based on +nown demands and optimi4e networ+ utili4ation 9asic allows primitive traffic engineering E '.g. dynamic routing ;70 ma+es networ+ resource commitment easy E ble to balance load in face of demand E ble to commit to different levels of support to meet user traffic re$uirements E ware of traffic flows with Jo0 re$uirements and predicted demand E ntelligent re)routing when congested #PN $upport &raffic from a given enterprise or group passes transparently through an internet
?5
0egregated from other traffic on internet erformance guarantees 0ecurity
Multiprotocol $upport ;70 can be used on different networ+ technologies E Re$uires router upgrades *oe-ist with ordinary routers &; &; E 'nables and ordinary switches co)e-ist Frame relay E 'nables and ordinary switches co)e-ist ;i-ed networ+ MPL$ Te Terminology rminology MPL$ @peration 7abel switched routers capable of switching and routing pac+ets based on label appended to pac+et be tween end points or multicast destinations 7abels define a flow of pac+ets between 'ach distinct flow %forward e$uivalence class E F'*( has specific path through 70Rs defined E *onnection oriented 'ach F'* has Jo0 re$uirements header not e-amined E Forward based on label value MPL$ @peration iagram
?=
2:planation 9 $etup 7abelled switched path established prior to routing and delivery of pac+ets Jo0 parameters established along path E Resource commitment E Jueuing and discard policy at 70R E nterior routing protocol e.g. 80F used E 7abels assigned 7ocal significance only ;anually or using 7abel distribution protocol %7D( or enhanced version of R0A 2:planation 9 Packet 0andling ac+et enters domain through edge 70R E rocessed to determine Jo0 70R assigns pac+et to F'* and hence 70 E ;ay need co)operation to set up new 70 ppend label Forward pac+et Within domain 70R receives pac+et Remove incoming label, attach outgoing label and forward 'gress edge strips label, reads header and forwards Notes
;70 domain is contiguous set of ;70 enabled enab led routers &raffic may enter or e-it via direct connection to ;70 router or from non);70 router F'* determined by parameters, e.g. E 0ource:destination address or networ+ address
?B
E ort numbers E protocol id E Differentiated services codepoint E v= flow label Forwarding is simple loo+up in predefined table E ;ap label to ne-t hop *an define <9 at an 70R for given F'* ac+ets between same end points may belong to different F'* MPL$ Packet Forwarding
La7el $tacking
ac+et may carry number of labels 7F8 %stac+( E rocessing based on top label E ny 70R may push or pop label 3nlimited levels E llows aggregation of 70s into single 70 for part of route E *.f. &; &; virtual virtual channels inside virtual paths E '.g. aggregate all enterprise traffic into one 70 for access provider to handleReduces si4e of tables
La7el Format iagram
Time to Li%e Processing since header not read !eeded to support &&7 since First label &&7 set to header &&7 on entry to ;70 domain &&7 of top entry on stac+ decremented at internal 70R E f 4ero, pac+et dropped or passed to ordinary error processing %e.g. *;( E f positive, value placed in &&7 of top top label on stac+ and pac+et forwarded t e-it from domain, %single stac+ entry( &&7 decremented E f 4ero, as above E f positive, placed in &&7 field of p header and La7el $tack ppear after data lin+ layer header, before networ+ layer header &op of stac+ is earliest %closest to networ+ layer header( &op !etwor+ layer pac+et follows label stac+ entry with 06 8ver connection oriented services E &opmost &opmost label value in & &; header A:A* field Facilitates &; switching E &op &op label inserted between cell header and header E n D7* field of Frame Relay E !ote" &&7 problem problem
??
Position o! MPL$ La7el $tack
F2"sD L$PsD and La7els &raffic grouped into F'*s &raffic in a F'* transits an ;70 domain along an 70 ac+ets identified by locally significant label t each 70R, labelled pac+ets forwarded on basis of label. E 70R replaces incoming label with outgoing label 'ach flow must be assigned to a F'* Routing protocol must determine topology and cu rrent conditions so 70 can be assigned to F'* E ;ust be able to gather and use information to support Jo0 70Rs must be aware of 70 for given F'*, assign incoming label to 70, communicate label to other 70Rs Topology o! L$Ps 3ni$ue ingress and egress 70R E 0ingle path through domain 3ni$ue egress, multiple ingress 70Rs E ;ultiple paths, possibly sharing final few hops ;ultiple egress 70Rs for unicast traffic ;ulticast
/oute $election 0election of 70 for particular F'*
?
"onstraint 'ased /outing Algorithm &a+e in to account traffic re$uirements of flows and resources available along hops E *urrent utili4ation, e-isting capacity, committed services E dditional metrics over and above traditional routing protocols %80F( ;a- lin+ data rate *urrent capacity reservation ac+et loss ratio 7in+ propagation delay
La7el istri7ution 0etting up 70 ssign label to 70 nform all potential upstream nodes of label assigned by 70R to F'* E llows proper pac+et labelling E 7earn ne-t hop for 70 and label that downstream node has assigned to F'* llow 70R to map incoming to outgoing label /eal Time Transport Protocol &* not suited to real time distributed application E oint to point so not suitable for multicast E Retransmitted segments arrive out of order E !o way to associate timing with segments 3D does not include timing information nor any support for real time applications 0olution is real)time transport protocol R& /TP Architecture *lose coupling between protocol and application layer functionality E Framewor+ for application to implement single protocol pplication level framing ntegrated layer processing
Application Le%el Framing Recovery of lost data done by application rather than transport layer E pplication may accept less than perfect delivery Real time audio and video nform source about $uality of delivery rather than retransmit 0ource can switch to lower $uality E pplication may provide data for retransmission 0ending application may recompute lost values rather than storing them
/
0ending application can provide revised values *an send new data to fi-Y conse$uences of loss 7ower layers deal with data in units provided by application E pplication data units %D3(
Integrated Layer Processing
dMacent layers in protocol stac+ tightly coupled llows out of order or parallel functions from different layers
/TP Architecture iagram
/TP ata Trans!er Protocol &ransport of real time data among number of participants in a session, defined b y" E R& ort number 3D destination port number if using 3D E R& *ontrol rotocol %R&*( port number Destination port address used by all participants for R&* transfer E addresses ;ulticast or set of unicast Multicast $upport 'ach R& data unit includes" 0ource identifier &imestamp ayload format /elays ntermediate system acting as receiver and transmitter for given protocol layer ;i-ers E Receives streams of R& pac+ets from one or more sources E *ombines streams E Forwards new stream &ranslators E roduce one or more outgoing R& pac+ets for each incoming pac+et E '.g. convert video to lower $uality /TP 0eader
/TP "ontrol Protocol (/T"P) R& is for user data R&* is multicast provision of feedbac+ to sources and session participants 3ses same underlying transport protocol %usually 3D( and different port number R&* pac+et issued periodically by each participant to other session members /T"P Functions Jo0 and congestion control dentification 0ession si4e estimation and scaling 0ession control
/T"P Transmission !umber of separate R&* pac+ets bundled in single 3D datagram E 0ender report E Receiver report E 0ource description E >oodbye E pplication specific /T"P Packet Formats
2
Packet Fields (All Packets) Aersion %2 bit( currently version 2 adding % bit( indicates padding bits at end of control information, with number of octets as last octet of padding *ount %5 bit( of reception report bloc+s in 0R or RR, or source items in 0D'0 or 9' ac+et type %? bit( 7ength %= bit( in 2 bit words minus n addition 0ender and receiver reports have" E 0ynchroni4ation 0ource dentifier Packet Fields ($ender /eport) $ender In!ormation 'lock !& timestamp" absolute wall cloc+ time when report sent R& &imestamp" Relative time used to create timestamps in R& pac+ets 0enderZs pac+et count %for this session( 0enderZs octet count %for this session( Packet Fields ($ender /eport) /eception /eport 'lock 00R*n %2 bit( identifies source refered to by this report bloc+ Fraction lost %? bits( since previous 0R or RR *umulative number of pac+ets lost %21 bit( during this session '-tended highest se$uence number received %2 bit( E 7east significant = bits is highest R& data se$uence number received from 00R*n