“Promoting the Use of End-to-End Congestion Control in the Internet”

04/19/2304/19/23 CS577 Spring 2005CS577 Spring 2005 11

““Promoting the Use of End-to-Promoting the Use of End-to-End Congestion Control in the End Congestion Control in the

Internet”Internet”Sally Floyd, Kevin Fall in Proceedings of Sally Floyd, Kevin Fall in Proceedings of IEEE/ACM Transactions on Networking, 1999IEEE/ACM Transactions on Networking, 1999

A Summary by Ashish Samant A Summary by Ashish Samant

CS577 – Spring 2005 CS577 – Spring 2005


OutlineOutline

Need for end-to-end congestion controlNeed for end-to-end congestion control Unfairness, Congestion CollapseUnfairness, Congestion Collapse Per flow based scheduling Vs Congestion Per flow based scheduling Vs Congestion

Control mechanisms at the routerControl mechanisms at the router Identifying candidate flows for regulationIdentifying candidate flows for regulation Other incentives for flowsOther incentives for flows


IntroductionIntroduction

End hosts/applications may not use end-End hosts/applications may not use end-to-end congestion control schemes.to-end congestion control schemes.

Problem - Problem - This may lead to congestion This may lead to congestion collapse and unfairness, in times of collapse and unfairness, in times of congestion.congestion.

Solution – Solution – Isolate ill-behaving flows, use Isolate ill-behaving flows, use per-flow based queuing at routers.per-flow based queuing at routers.

This may not be sufficient !! This may not be sufficient !!


Introduction …. continuedIntroduction …. continued

Authors suggest - Authors suggest -

Routers must support congestion control Routers must support congestion control and regulate high-bandwidth flows.and regulate high-bandwidth flows.

Routers must regulate ‘best effort flows’ thatRouters must regulate ‘best effort flows’ that

are TCP-Unfriendly,are TCP-Unfriendly,unresponsive to congestion,unresponsive to congestion,use disproportionate bandwidth.use disproportionate bandwidth.


Introduction …. continuedIntroduction …. continued

Unresponsive flows cause two problemsUnresponsive flows cause two problems - - Unfairness; well-behaved flows may Unfairness; well-behaved flows may suffer bandwidth starvation because suffer bandwidth starvation because unresponsive flows do not react to unresponsive flows do not react to congestion.congestion.

- Congestion collapse; the scarce - Congestion collapse; the scarce bandwidth bandwidth of the network is consumed of the network is consumed by by packets from unresponsive flows, that packets from unresponsive flows, that

will be discarded sooner or later.will be discarded sooner or later.


Experimental SetupExperimental Setup


Unfairness – 3 TCP, 1 UDP flow, Unfairness – 3 TCP, 1 UDP flow, FCFSFCFS


Fairness – 3 TCP, 1 UDP flow, WRRFairness – 3 TCP, 1 UDP flow, WRR


Congestion Collapse – 3 TCP, 1 Congestion Collapse – 3 TCP, 1 UDP flow, FCFSUDP flow, FCFS


Congestion Control – 3 TCP, 1 UDP Congestion Control – 3 TCP, 1 UDP flow, WRRflow, WRR


Congestion Control – 3 UDP, 1 TCP Congestion Control – 3 UDP, 1 TCP flow, WRRflow, WRR


Identifying non TCP-Friendly FlowsIdentifying non TCP-Friendly Flows

TCP Friendly Flow – arrival rate does not TCP Friendly Flow – arrival rate does not exceed that of any other TCP conformant flow.exceed that of any other TCP conformant flow.

Maximum sending rate for a TCP Friendly Maximum sending rate for a TCP Friendly flow :flow :

T - sending rate ; p - packet drop rate ; B – max packet size ; T - sending rate ; p - packet drop rate ; B – max packet size ; R – minimum RTTR – minimum RTT

Actual rates will be less than T.Actual rates will be less than T.


Identifying non TCP-Friendly FlowsIdentifying non TCP-Friendly Flows

Limitations : Limitations : Inconsistencies in finding packet size, round Inconsistencies in finding packet size, round

trip trip time.time.

Measurements should be taken over a long Measurements should be taken over a long interval of time.interval of time.

Bursty packet drops.Bursty packet drops.

Router Response :Router Response :Routers should ‘freely restrict’ the bandwidth of Routers should ‘freely restrict’ the bandwidth of

non non

TCP – Friendly flows.TCP – Friendly flows.


Identifying Unresponsive FlowsIdentifying Unresponsive Flows

TCP Friendly test cannot be used at TCP Friendly test cannot be used at routers that are unable to determine routers that are unable to determine packet sizes and RTTs.packet sizes and RTTs.

If packet drop rates increase by If packet drop rates increase by x x , , the arrival rate should drop by the arrival rate should drop by √√xx . .

When packet drop is constant, no When packet drop is constant, no flow will be identified as flow will be identified as unresponsive. unresponsive.


Identifying Unresponsive FlowsIdentifying Unresponsive Flows

Limitations : Limitations : Packet drop may be because of various Packet drop may be because of various

reasons, reasons, hard for flows with variable demand.hard for flows with variable demand.Flows might be tempted to start with a Flows might be tempted to start with a

higher initial higher initial bandwidth demand.bandwidth demand. Response :Response :

Actively regulate the bandwidth of Actively regulate the bandwidth of unresponsive unresponsive flows in times of congestion.flows in times of congestion.


Identifying flows using Identifying flows using disproportionate flowsdisproportionate flows

Flows that require larger bandwidth than other Flows that require larger bandwidth than other flows that reduce their demand.flows that reduce their demand.

These might be TCP friendly but still be These might be TCP friendly but still be ‘disproportionate’.‘disproportionate’.

Arrival rate <= log(3n) / n ; Arrival rate <= log(3n) / n ; n = no of flowsn = no of flows

Arrival rate <= c / Arrival rate <= c / √p ; √p ; p = pkt drop rate p = pkt drop rate c = some constantc = some constant


Comments and ConclusionComments and Conclusion

Alternate approaches Alternate approaches - use schemes that are a mix of FCFS and per-flow based - use schemes that are a mix of FCFS and per-flow based approach ( FCFS scheduling with differential dropping ).approach ( FCFS scheduling with differential dropping ).- pricing incentives.- pricing incentives.

granularity of flowsgranularity of flows- apply fairness tests to single/aggregate of flows.- apply fairness tests to single/aggregate of flows.

min-max fairness measure min-max fairness measure - need to look at the entire path, all the congested links.- need to look at the entire path, all the congested links.


Comments and Conclusion … Comments and Conclusion … continuedcontinued

Breaking a TCP connection, Breaking a TCP connection, increased local throughput but also increased local throughput but also increases global packet drop rate.increases global packet drop rate.


Derivation of TCP Friendly RateDerivation of TCP Friendly Rate

Once congestionWindow >= W ; a packet is Once congestionWindow >= W ; a packet is dropped and the congestion window is halved. dropped and the congestion window is halved.

As long as congestionWindow < W ; window is As long as congestionWindow < W ; window is increased by 1, per RTTincreased by 1, per RTT

W/2 + (W/2+1) + (W/2+2) + … W = 3/8*WW/2 + (W/2+1) + (W/2+2) + … W = 3/8*W2 2

=> per packet drop ; max 3/8*W=> per packet drop ; max 3/8*W2 2 packets are sent packets are sent

=> max packet drops <= 1/(3/8*W=> max packet drops <= 1/(3/8*W22))


Derivation of TCP Friendly Rate … Derivation of TCP Friendly Rate … ContinuedContinued

Max bytes transferred per cycle of steady state:Max bytes transferred per cycle of steady state:

Total packets sent * Avg. packet Size Total packets sent * Avg. packet Size

Avg Round Trip TimeAvg Round Trip Time

( Total packets sent = 0.75*W )( Total packets sent = 0.75*W )

= > (0.75 * W * B) / R = > (0.75 * W * B) / R

= > = >

Documents

“Promoting the Use of End-to-End Congestion Control in the Internet”