Texas A&M University {sumitha,saurabhj,reddy}@ee.tamu.edu Improving TCP Performance in High Bandwidth High RTT Links Using Layered Congestion Control Sumitha.

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Improving TCP Performance in High Bandwidth High RTT Links Using Layered Congestion Control

Sumitha Bhandarkar Saurabh Jain

A. L. Narasimha ReddyTexas A & M University

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Layering Concepts

• Design Constraints – Fairness among flows of similar RTT – RTT unfairness no worse than TCP– Fair to TCP in slow networks

• Two dimensional congestion control– Increase layers, if no losses for extended period– Per-RTT window increase more aggressive at

higher layers

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• Layering– Start layering when window > WT

– Associate each layer with a step size K

– When window increases from previous addition of layer by K, increment number of layers

– For each layer K, increase window by K per RTT

Number of layers determined dynamically based on current network conditions.

Layering Concepts (Cont.)

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

K

Layering Concepts (Cont.)

K + 1

K

K - 1

LayerNumber

WK-1

Minimum Window Corresponding to the layer

Number of layers = K when WK W WK+1

WK

WK+1

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• Constraint 1 : – rate of increase for flow at higher layer should be lower

than flow at lower layer

• Constraint 2 : – After a loss, recovery time for a larger flow should be more

than the smaller flow

(K1 > K2, for all K1, K2 2)

Framework

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• Decrease behavior : – Multiplicative decrease

• Increase behavior :– Additive increase with additive factor = layer

number

W = W + K/W

A Design Choice

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• After loss, drop at most one layer

• Constraint for choice of K:

• We choose

A Design Choice (Cont.)

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• Choice of :Since after loss, at most one layer is dropped,

(We choose = 0.15 corresponding to K = 19)

A Design Choice (Cont.)

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Time to claim bandwidth

Analysis

Speedup inPacket recovery time

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• Steady state throughput

where K' is the layer corresponding to steady state window size, is the window decrease factor and p is the steady state loss probability

Analysis (Cont.)

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• RTT Unfairness– With random losses, RTT unfairness similar to TCP– With synchronized losses, RTT unfairness is– Can be easily compensated

• Modify increase behavior W = W + (KR * K) / W

• When KR RTT (1/3), RTT unfairness similar to TCP

• When KR RTT, linear RTT unfairness (window size independent of RTT)

– Loss model depends on type of queue management, level of multiplexing etc.

Analysis (Cont.)

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Window Comparison

Experimental Evaluation

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Link Utilization

Experimental Evaluation

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Fairness among multiple flows

Experimental Evaluation

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Dynamic Link Sharing

Experimental Evaluation

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Interaction with TCP

Experimental Evaluation

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

RTT Unfairness

Experimental Evaluation

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• Why LTCP ?– Current design remains AIMD– Dynamically changes increase factor– Retains convergence and fairness properties– Simple to understand/implement– RTT unfairness similar to TCP

Conclusions

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• Characterize losses on actual high speed links

• Study alternate designs for LTCP framework

• Compare with other TCP based high speed solution. Preliminary results show – observed loss probability with LTCP is lower

than other schemes– improved RTT unfairness– better TCP tolerance in high speed networks

Future Work

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

RTT Unfairness

Comparison with BIC(Preliminary Results)

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Questions ?

Additional questions/feedback welcome at

{sumitha,saurabhj,reddy}@ee.tamu.edu

Thank You...

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Simulation Topology

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• HS-TCPSally Floyd, “HighSpeed TCP for Large Congestion Windows”, RFC 3649 Dec 2003.

• Scalable TCPTom Kelly, “Scalable TCP: Improving Performance in HighSpeed Wide AreaNetworks”, ACM Computer Communications Review, April 2003.

• FASTCheng Jin, David X. Wei and Steven H. Low, “FAST TCP: motivation, architecture,algorithms, performance”, IEEE Infocom, March 2004.

• BICLisong Xu, Khaled Harfoush, and Injong Rhee, “Binary Increase Congestion Control forFast Long-Distance Networks”, IEEE Infocom, March 2004.

• HTCPR. N. Shorten, D. J. Leith, J. Foy, and R. Kilduff, “H-TCP Protocol for High-Speed LongDistance Networks”, PFLDnet 2004, February 2003.

Related Work

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

• Probability of loss for LTCP

• Probability of loss for TCP

RTT Fairness(Random Loss Model)

Texas A&M University{sumitha,saurabhj,reddy}@ee.tamu.edu

Observed Loss Rates

Comparison with BIC(Preliminary Results)

Single Flow, 1Gbps bottleneck link