Solutions for Improving TCP Performance over Emerging Networks

1
Solutions for Improving TCP Performance over
Emerging Networks

• K.R.Renjish Kumar
• kaleelaz_at_comp
• Centre for Internet Research

2
Outline

• Introduction
• Emerging Networks
• TCP
• TCP over Emerging Networks
• Issues
• Solutions
• TCP HACK
• MBM

3
Introduction

• Internet has come a long way with
• Different kinds of networks
• increasing number of users
• innumerable applications which require better
  quality of service.

4
Emerging Networks

• Lossy Link networks
• Wireless networks like wavelans
• Satellite networks
• Quality of Service(QoS) model based networks
• Integrated Services (IntServ) Provides QoS in a
  per connection basis
• Differentiated Services (DiffServ) Provides QoS
  for an aggregate flow. Much more scalable
  compared to IntServ.

5
Transmission Control Protocol (TCP)

• De facto Transport layer Protocol
• 95 of data byte share in the Internet
• Common applications like Telnet,FTP and HTTP uses
  TCP

6
TCP contd

• TCP Header
• Sequence Number the first byte of data in this
  segment.
• Options For any additional features.
• Rate Control
• Slow start,fast retransmit,recovery,
• congestion avoidance.
• Segments send equal to the minimum of the
  senders window(cwnd) and receivers window.

7
Emerging Networks-Lossy Networks

• Access Networks are increasingly becoming
  wireless eg wavelans
• Acts as bottlenecks. Bandwidth limited.
• Prone to higher rate of corruption

8
TCP over Lossy Links

• Issues
• performs poorly in lossy links where the rate of
  corruption of packets are higher
• Reason TCP assumes that loss of packets are due
  to congestion and not corruption, thus reducing
  the cwnd and thus reducing the throughput.

9
Solution

• TCP HACK TCP Header Checksum Option to Improve
  Performance over Lossy Links.
• IEEE INFOCOM 2001,Anchorage,Alaska
• R.K.Balan,B.P.Lee,K.R.R.Kumar,L.Jacob,W.K.G.
  Seah,
• A.L.Ananda, IEEE INFOCOM
  2001,Anchorage,Alaska

10
TCP HACK

• Key Observations
• For most data transfers, the data portion of a
  TCP packet is much larger than the header portion
  of the packet (may not be true for interactive
  traffic)
• As such, corruptions are far more likely to occur
  in the data portion and not the header portion
• Given this scenario, can we make use of the
  intact header to make TCP perform better?

11
TCP Hack Algorithm contd..

• On receiving this specially marked ACK packet,
  the sender will know that packet corruption and
  not congestion has occurred
• As such, the sender will retransmit the corrupted
  packet and will not half its congestion window

12
Header Checksum option
TCP Header Checksum option
TCP Header Checksum ACK option
13
TCP Sender
14
TCP Receiver
Data segment received
No
TCP segment corrupted?
Continue as per normal
Yes
Header portion corrupted?
Yes
Discard Packet
No

• ) Recover sequence number of corrupted segment
  from header.
• ) Generate special ACK (option 15) containing
  the sequence
• number of the corrupted segment.

15
ACK Processing
ACK segment received
No
Option 15 present in the segment?
Yes
Modification to the ACK processing
16
Note

• For TCP HACK to be effective, link layer
  retransmissions and error recovery should be
  turned off
• Corrupted packets should be sent up to the TCP
  layer where TCP HACK will do the retransmissions

17
Experimental Setup

• Implemented into the Linux 2.2.10 kernel
• Test bed was set up comprising of 3 machines
• Client and server were running TCP HACK
• 2 different models were used
• Single packet corruption.
• Burst corruption with differing burst lengths

18
Low Latency Links
19
High Latency Links
20
Number of slow starts
21
Observations

• Both SACK and TCP-HACK are better than NewReno
• Hence NewReno was removed from the tests when
  burst errors were introduced

22
Burst errors
Throughput for 2 burst error for various burst
lengths
Throughput for 5 burst error for various burst
lengths
23
Throughput vs Time
24
Inference

• How is TCP HACK doing retransmissions a good
  idea?
• Could help stabilize TCP and prevent its RTT
  calculation from fluctuating widely (especially
  useful when using a long latency link)
• Provides more information to TCP, thus preventing
  it from unnecessarily timing out
• Distinguishes between congestion and corruption
• Improves overall TCP performance

25
Emerging QoS Network Architecture

• Differentiated Services (DiffServ)
• Provides QoS for aggregate flows using different
  per hop forwarding behaviours at the core
  routers.
• Basic building blocks of DiffServ
• Classifiers
• Meters
• Markers
• Policers

26
DiffServ
Meter
Packet
Classifier
Marker
Policer
Logical View of a Packet Classifier and Traffic
Conditioner
27
TCP over QoS based Networks

• Issues
• Bias against connections with long Round Trip
  Time (RTT)
• ReasonLong RTT flows takes longer time to ramp
  up.
• Bias against connections with smaller window
  sizes
• Reason Smaller windows mean smaller throughput
• Protection from unruly traffic like UDP traffic.
• Reason UDP traffic has no rate control mechanism
  and hence kills TCP traffic.

28
Solution

• A Memory-Based Approach for a TCP-Friendly
  Traffic Conditioner in DiffServ Networks.to
  appear in Proc. of the 9th IEEE International
  Conference on Network Protocols (ICNP 2001),
  Riverside, California. 
• K.R.R.Kumar, A.L.Ananda, Lillykutty
  Jacob

29
Topology
30
Memory-Based Marker (MBM)

• Key observations
• Markers, one of the building blocks of a traffic
• conditioner play a major role for resource
  allocation
• in a DiffServ network.
• TCP performance is highly influenced by two
• parameters, namely RTT, and window size.
• Markers are mostly deployed at the edge routers,
• which cannot easily decide the window size and
  RTT of the various TCP connections passing
  through.

31
What is needed?

• Marking algorithm
• Least sensitive to both the marker and TCP
  parameters
• Should be transparent to end hosts.
• Maintain optimum marking
• Tracks the TCP dynamics

32
Why memory-based marker?

• During the period when TCP flows experience
• congestion, either or both of the following
  occurs
• a) The cwnd reduces reducing the value of W
• b) The RTT increases
• causing a decrease in throughput or rate of
  flow.

33
Why memory-based marker?

• The TCP window size W and the round trip time RTT
  are related to the throughput by the equation
• BW ¾(MSSW)/(RTT) where W is expressed in
• number of segments.
• Any variation in W or RTT is reflected as
  subsequent changes in BW, i.e., in our case, the
  avg_rate.
• The parameter previous average rate (par) is
  compared with the present average rate to track
  any change in the rate of flow and thus
  indirectly extract the variations in RTT or W.

34
MBM Algo. explained

• In the expression for the marking probability mp,
• (par avg?rate)/avg?rate tracks the variations
  in RTT and window size (W) and thus increases or
  decreases the marking probability according to
  the changes in the flow rate.
• (1- avg?rate)/cir constantly compares the
  average rate observed with the target rate to
  keep the rate closer to the target.

35
MBM Algorithm

• For each packet arrival
• If avg_rate ? cir
• then
• mpmp(1-avg_rate/cir)(par-
  avg_rate)/avg_rate
• par avg_rate
• mark the packet using
• cp 11 w.p. mp
• cp 00 w.p. (1-mp)

• else if avg_rate gt cir
• then
• mp mp (par avg_rate)/avg_rate
• paravg_rate
• mark the packet using
• cp 11 w.p. mp
• cp 00 w.p. (1-mp)

36
MBM Algo. Contd

• where,
• avg_rate the rate estimate upon each packet
  arrival
• mp marking probability (?1)
• cir committed information rate (i.e., the
  target rate)
• par previous average rate
• cp denotes codepoint and w.p. denotes with
• probability.

37
Experiments

• We used FTP bulk data transfer for the TCP
  traffic in all our experiments.
• We conducted simulation studies in NS for
• Assured service for aggregates with different
  target rates.
• Effect of different RTTs
• Effect of different window sizes
• Protection from best effort UDP flows
• Effect of UDP flows with target rates.

38
Results
Achieved Rates (Ra) for different Target Rates
(Rt).
39
Results.
Achieved Rates (Ra) for different RTT values
Achieved Rates (Ra) for different window sizes
40
More results
Achieved Rates in presence of BE UDP and TCP
Achieved Rates in presence of AS UDP and BE TCP
41
Inference

• MBM
• Achieves transparency from the end hosts,
  simplicity, and least sensitivity to parameters
  of both TCP as well as its own parameters.
• helps in achieving the target rate, with a better
  fairness in terms of sharing the excess bandwidth
  among flows.
• provides the TCP flows, a greater degree of
  insulation
• from differences in RTT and window sizes.
• The overall link utilization also seems to be
  much better.

42
Future work

• Working on a 3 colour marker, Memory-Based Three
  color Marker (MBTCM) which is apt for DiffServ
  with AF per hop behaviour in its core.
• Study of TCP dynamics and its performance
  improvements over broadband networks.

43
Related Links..

• http//www.comp.nus.edu.sg/srijith/cir/papers.htm
  l
• http//www.comp.nus.edu.sg/kaleelaz