on tcp-jersey - wocc · 2007-06-15 · 1 advanced networking laboratory nirwan ansari advanced...

18
1 Advanced Networking Laboratory Nirwan Ansari Advanced Networking Laboratory Department of Electrical and Computer Engineering New Jersey Institute of Technology Newark, NJ 07102-1982, USA Phone: +1-973-596-3670 Email: [email protected] http://web.njit.edu/~ansari On TCP-Jersey Advanced Networking Laboratory © 2007 Nirwan Ansari Acknowledgements Kai Xu Ye Tian

Upload: others

Post on 14-Mar-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

1

Advanced Networking Laboratory

Nirwan AnsariAdvanced Networking Laboratory

Department of Electrical and Computer EngineeringNew Jersey Institute of Technology

Newark, NJ 07102-1982, USAPhone: +1-973-596-3670

Email: [email protected]://web.njit.edu/~ansari

On TCP-Jersey

Advanced Networking Laboratory © 2007 Nirwan Ansari

AcknowledgementsKai XuYe Tian

2

Advanced Networking Laboratory © 2007 Nirwan Ansari

Outline

Brief Overview of TCP Congestion Control

Challenges to TCP Performance in Wireless Networks

Design Goals and Solution Framework

TCP-Jersey

Conclusions

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP Congestion ControlSliding & Elastic Window-Based Congestion Control

cwnd determines the number of allowable unacknowledged pkts in network

cwnd advancement clocked by arrivals of ACKs, i.e., ACK-clocking

ACKs are cumulative, i.e., asking for the next expected pkt

Packet loss results in duplicated ACKs (DUPACK)

Two main phasesSlow Start: coarse-grain probing of the equilibrium, cwnd grows exponentially

Congestion Avoidance: AIMD

fine-grain probing of the equilibrium, cwnd grows additively

cwnd cut by half on 3 DUPACKs, multiplicative rate decrement (Reno)

All packet losses are interpreted as network congestion

3

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP Congestion Control (cont’d)Typical cwnd evolution of TCP-Tahoe Typical cwnd evolution of TCP-Reno

Advanced Networking Laboratory © 2007 Nirwan Ansari

Outline

Brief Overview of TCP Congestion Control

Challenges to TCP Performance in Wireless Networks

Design Goals and Solution Framework

TCP-Jersey

Conclusions

4

Advanced Networking Laboratory © 2007 Nirwan Ansari

Challenges to TCP Performance in Wireless Networks

Wireless links vs. wired links:

High Bit Error Rate (BER)

Air vs. copper/fiber as transmission medium

Interferences (microwave oven & cordless phone v. 802.11b/g)

Signal fading (distance, multi-path reflections, etc.)

Weather conditions (lightening, etc.)

Frequent disconnection/reconnection

Moving obstacles (trucks, trains, etc.)

User mobility (hand-off, passing through building, etc.)

Advanced Networking Laboratory © 2007 Nirwan Ansari

On wireless links: random & busty packet losses may not be

congestion related.

Result in TCP performance:

unnecessary rate reduction,

hence, throughput degradation,

low link utilization,

unstable buffer occupancy,

volatile end-to-end delay/jitter

Loss differentiation at transport layer is hard:

Congestion induced packet losses can appear to be random and bursty as

well

Challenges to TCP Performance in Wireless Networks

Because TCP does not differentiate loss types

5

Advanced Networking Laboratory © 2007 Nirwan Ansari

Examples of TCP Performance Degradation over Lossy Links

Transmission back-off due to time-out caused by short link disconnection

Advanced Networking Laboratory © 2007 Nirwan Ansari

Performance of Standard TCP in Wireless Networks

Reno model

( )20

1( ) min , 2 3min 1, 3 1 32

3 8

mwB pRTT bp bpRTT T p p

⎧ ⎫⎪ ⎪⎪ ⎪≈ ⎨ ⎬⎧ ⎫⎪ ⎪+ +⎨ ⎬⎪ ⎪⎩ ⎭⎩ ⎭

Simulations

Random packet losses

S D

100Mbps,45ms 5Mbps,1ms

BS

6

Advanced Networking Laboratory © 2007 Nirwan Ansari

Outline

Brief Overview of TCP Congestion Control

Challenges to TCP Performance in Wireless Networks

Design Goals and Solution Framework

TCP-Jersey

Conclusions

Advanced Networking Laboratory © 2007 Nirwan Ansari

Design GoalsEvolutionary not Revolutionary (for the time-being)

Compatibility

No application code changes

Coexist & inter-operable with standard TCP stacks

Transparent to lower-layer functions, e.g., FEC, ARQ, etc.

End-to-end

Congestion control and loss differentiation done at transport layer

No “wireless-aware” routers,

No cross-layer signaling

Simplicity, scalability, low implementation overhead

7

Advanced Networking Laboratory © 2007 Nirwan Ansari

Framework of Proposed SolutionApproach the problem in two inter-related fronts:

Congestion Control: More sensible and adaptive window reduction scheme;

rate halving is too drastic, particularly unsuitable over wireless links.

But to which level does the sender reduce transmission rate?

Loss Differentiation: So that losses due to wireless errors are treated

differently than congestion induced losses.

But by what mechanism does the protocol distinguish different causes of packet

loss at the transport layer?

Advanced Networking Laboratory © 2007 Nirwan Ansari

Achievable Rate Estimation (ARE) Based AIAD Congestion Control, cut

cwnd to AR instead of by half, thus adaptive to achievable rate

AR: sustainable rate w/o causing congestion. Makes more sense

Robust to undifferentiated random packet losses

Framework: Adaptive Congestion Control

, im i iiTw Bµ= +A better congestion control alone can be effective to random losses, but can be done better.

8

Advanced Networking Laboratory © 2007 Nirwan Ansari

Framework: Loss Differentiation via AQMLoss Differentiation via Congestion Warning (CW) Packet Marking

leverage the ECN functionality at routers

CW mark in DUPACK tells loss type (congestion v. wireless). No cwnd

reduction if loss is attributed to wireless link errors

Advanced Networking Laboratory © 2007 Nirwan Ansari

Outline

Brief Overview of TCP Congestion Control

Challenges to TCP Performance in Wireless Networks

Design Goals and Solution Framework

TCP-Jersey

Conclusions

9

Advanced Networking Laboratory © 2007 Nirwan Ansari

Essence of TCP-Jersey: an integrated implementation

1. Adaptive congestion control at source

Guided by timestamps-based achievable rate estimation (ARE), i.e., the rate sustainable without causing congestion

ARE is immune to reverse path congestion, capacity asymmetry, packet loss on reverse path

Never halves the window

2. Congestion warning packet marking at link

Compatible to ECN and the like

Serves two purposes, i.e., congestion signaling and helping in loss type inference

3. Packet loss type differentiation at source

Analyzing the CW mark in DUPACK, and acting accordingly

Simple and effective

Advanced Networking Laboratory © 2007 Nirwan Ansari

Achievable Rate Estimation (ARE) at Source

( )1

1

k kk

k k

k

k

R LRt tτ

τ−

× +=

− +

ARE via Returning ACKs: a modified TSW with dynamic time constant

kL Bytes acknowledged by kth ACK

1k k kt t −∆ ≡ − Inter-arrival of ACKs

kτSum of RTT and RTT variation (changes with network conditions)

kk

k

Lr =∆

Instantaneous rate sample at kth ACK arrival (the input to the filter)

Implementation: sender updates the estimate upon each ACK arrival

10

Advanced Networking Laboratory © 2007 Nirwan Ansari

Congestion Warning (CW) Packet Marking at Link (AQM)

Receiver echoes the mark in TCP header

Router signals congestion by marking in IP header

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP-Jersey: The Algorithm

if CW mark is set in DUPACK {

lost pkt may due to congestion

retransmit lost pkt

reduce cwnd to ARExD

}

else {

lost pkt may due to error

retransmit lost pkt

keep cwnd intact

}

lost pkt may due to congestion

retransmit lost pkt

reduce cwnd by 1/2

TCP-Jersey Stock TCP

Reaction to Loss Detected via DUPACK

11

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP-Jersey: The Algorithm (cont’d)

if in congestion state {

lost pkt may due to congestion

retransmit lost pkt

reduce cwnd to 1

}

else {

lost pkt may due to error

retransmit lost pkt

reduce cwnd to ARExD

}

lost pkt may due to congestion

retransmit lost pkt

reduce cwnd to 1

TCP-Jersey Stock TCP

Reaction to Loss Detected via TIMEOUT

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP-Jersey: The Algorithm (cont’d)

adjust cwnd to ARExD reduce cwnd by 1/2

TCP-Jersey Stock TCP

Reaction to CW mark in Normal ACK

12

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP-Jersey: Performance Evaluation -- Goodput

Case 1: Single TCP flow without congestion.

Wireless link has variable random packet loss rates ranging from 0.001% to 10%

Random packet losses

At 1% random packet loss rate,

TCP-Jersey outperforms standard Reno by 85% in goodput.

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP-Jersey: Performance Evaluation -- Goodput

Case 2: Same network. Single TCP flow with congestion inflicted by non-congestion controlled VoIP flows.

VoIP calls: Poisson arrival, by UDP at 96Kbps rate, max. 5 concurrent calls,

VoIP flows: i.i.d. exponential on-off with 50% talk spur & 50% silence period,

VoIP call duration: exponential distribution with average of 8 sec.

At 1% random packet loss rate,

TCP-Jersey outperforms standard Reno by 76% in goodput.

13

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP-Jersey: Performance Evaluation -- Fairness

0.970.970.9710.00.970.990.995.00.991.00.991.00.990.991.00.51.00.970.990.11.00.990.980.0

TCP-JerseyTCP-WestwoodTCP-RenoError Rate (%)

( )2

2i

i

xN x

φ = ∑∑

Random packet losses

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP-Jersey: Performance Evaluation -- Friendliness

m standard TCP-Reno flows and n TCP-Jersey flows sharing the same network

m+n is constant (20)

Case 1: wireless link is error-free

Case 2: wireless link has 0.1% of random packet loss rate

Compare average throughputs attained by Reno and Jersey flows

Random packet losses

14

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP-Jersey: Performance Evaluation -- Friendliness

959.40959.16317959.16959.21515958.97959.431010958.95959.94155958.79961.50173

Mean Jersey ThruMean Reno Thru# Jersey# Reno

1015.16944.353171007.40937.57515993.86916.201010972.33903.53155965.76895.23173

Mean Jersey ThruMean Reno Thru# Jersey# Reno

Over Error-Free Wireless Link

Over Lossy Wireless Link (0.1% packet loss rate)

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP-Jersey: Performance Evaluation – Co-Existing

446.40419.49177.20213.765.01298.091077.56554.78552.011.01455.361338.91936.60761.280.51623.641354.501541.921158.940.11663.591663.981659.361664.220.0

Jersey Throughput

Westwood Throughput

Vegas Throughput

Reno Throughput

Error Rate (%)

Random packet losses

15

Advanced Networking Laboratory © 2007 Nirwan Ansari

Outline

Brief Overview of TCP Congestion Control

Challenges to TCP Performance in Wireless Networks

Design Goals and Solution Framework

TCP-Jersey

Conclusions

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP Jersey --- SummaryTCP Jersey only requires simple configurations at the ECN-enabled route and minimum changes to the TCP sender side code without altering the protocol itself.Effectiveness of TCP Jersey over High Speed Downlink Packet Access (HSDPA) in the Universal Mobile Telecommunication System (UMTS)Further enhancements through better ARE and CW

16

Advanced Networking Laboratory © 2007 Nirwan Ansari

SummaryThe characteristics of wireless networks vary from applications to applications, and therefore, a universal solution for all kinds of wireless networks is unlikely to be developed.Nevertheless, the proactive TCP schemes have attracted much attention recently. They address the generalized issue of the heterogeneous network, i.e., the frequent random errors in the wireless network.To differentiate the wireless loss from congestive loss is the primary goal for proactive schemes.More importantly, proactive schemes are able to manage and utilize the available bandwidth in the network more efficiently.

Advanced Networking Laboratory © 2007 Nirwan Ansari

Summary (cont’d)Various proposals continue to emerge.Readers are referred to Reference [19] and [20] for further studies.

17

Advanced Networking Laboratory © 2007 Nirwan Ansari

TCP-Jersey NS codes (latest version) • http://web.njit.edu/anl/download.html

Advanced Networking Laboratory © 2007 Nirwan Ansari

References[1] J. Postel, “Internet protocol,” RFC 791, September 1981.[2] J. Postel, “Transmission control protocol,” RFC 793, September 1981.[3] K. Fall and S. Floyd, “Simulation-based comparisons of Tahoe, Reno and SACK TCP,”

Computer Communication Review, July 1996. [4] D. Chiu and R. Jain, “Analysis of the increase/decrease algorithms for congestion

avoidance in computer networks,” Journal of Computer Networks, vol. 17, no. 1, pp. 1–14, June 1989.

[5] I. F. Akyildiz, G. Morabito, and S. Palazzo, “TCP-Peach: a new congestion control scheme for satellite IP networks,” IEEE/ACM Trans. on Networking, vol. 9, no. 3, pp. 307–321, June 2001.

[6] J. Liu and S. Singh, “ATCP: TCP for mobile ad hoc networks”, IEEE JSAC, vol. 19, no. 7, pp. 1300-1315, 2001.

[7] T. Go , J. Moronski, D. S. Phatak, and Gupta, “Freeze-TCP: A true end-to-end enhancement mechanism for mobile environments,” Proc. INFOCOM 2000, 2000.

[8] A. Bakre, B. R. Badrinath, “I-TCP: Indirect TCP for mobile ohsts,” Proceedings of ICDCS 95, pp. 136–143, May 1995.

[9] S. Floyd and T. Henderson, “The new Reno modification to TCP’s fast recovery algorithm,” RFC2582, April 1999.

[10] M. Mathis, J. Mahdavi, S. Floyd, and A. Romanow, “TCP selective acknowledgment options,” RFC2018, October 1996.

[11] L. Brakmo and L. Peterson, “TCP Vegas: End to end congestion avoidance on a global Internet,” IEEE JSAC, Vol 13, No. 8, pp. 1465-1480, October 1995.

18

Advanced Networking Laboratory © 2007 Nirwan Ansari

References (cont’d)[12] C. P. Fu and S. C. Liew, “TCP Veno: TCP enhancement for transmission over wireless

access networks,” IEEE JSAC, vol. 21, no. 2, pp. 216–228, February 2004.[13] C. Casetti, M. Gerla, S. Mascolo, M. Y. Sanadidi, and R. Wang, “TCP Westwood:

bandwidth estimation for enhanced transport over wireless Links,” ACM Mobicom, pp. 287–297, July 2001.

[14] K. Xu, Y. Tian, and N. Ansari, “TCP-Jersey for wireless IP communications,” IEEE JSAC, vol. 22, no.4, pp. 747–756, May 2004.

[15] S. Li and N. Ansari, “TCP-Jersey over high speed downlink packet access,” Proc. IEEE GLOBECOM'05, St. Louis, MO, USA, Nov. 28 - Dec. 2, 2005, pp. 3576-3580.

[16] K. Xu, Y. Tian, and N. Ansari, "Improving TCP performance in integrated wireless communications networks," Computer Networks, Vol. 47, No. 2, pp. 219-237, February 2005.

[17] T. Taleb, N. Kato, and Y. Nemoto, “REFWA: An efficient and fair congestion control scheme for LEO satellite networks,” IEEE/ACM Trans. on Networking, Vol. 14, No. 5, pp. 1031-1044, October 2006.

[18] A. Arulambalam, X. Chen and N. Ansari, "Allocating Fair Rates for Available Bit Rate Services in ATM Networks," IEEE Communications Magazine, vol. 34, no. 11, pp. 92-100, Nov. 1996.

[19] Y. Tian, K. Xu, and N. Ansari, “TCP in wireless environments: Problems and solutions,”IEEE (Radio) Communications Magazine, vol. 43, no. 3, pp. S27-S32, March 2005.

[20] K.-C. Leung and V.O.K. Li, “Transmission Control Protocol (TCP) in Wireless Networks: Issues, Approaches, and Challenges,” IEEE Communications Surveys and Tutorials, Vol. 8, No. 4, 4th Quarter, 2006.

Advanced Networking Laboratory © 2007 Nirwan Ansari

Questions ?