Impact of Layer Two ARQ on TCP Performance in W-CDMA Networks

1
Impact of Layer Two ARQ on TCP Performance in
W-CDMA Networks

• Hiroshi Inamura, Osamu Takahashi, Hirotaka
  Nakano, Taro Ishikawa, NTT DoCoMo, Inc.
• Hiroshi Shigeno, Department of Information and
  Computer Science, Keio University
• Presented by Ming-Yen Lai

2
Outline

• Introduction
• Overviews WCDMA Network Features
• Related Works
• TCP Performance with RLC Retransmission
• Performance Evaluation via Emulation and
  Simulation
• Conclusion

3
Introduction

• The quality of transmission in the wireless
  networks that incur bit error can be improved
  with link layer retransmission.
• This paper addresses the interaction between TCP
  and RLC (Radio Link Control) in evaluating TCP
  performance.
• The paper examines the trade-off between link
  utilization and transport layer performance by
  changing the link control parameters.

4
Overviews WCDMA Network Features (cont.)

• WCDMA, an IMT2000 was developed by 3GPP.
• WCDMA has two types of services voice service on
  switched circuits and data service created by
  packet transfer.
• In packet service, WCDMA system controls
  transmission error by Radio Link Control (RLC).

5
Overviews WCDMA Network Features (cont.)

• WCDMA protocol stack

6
Overviews WCDMA Network Features (cont.)

• Retransmission in RLC

7
Overviews WCDMA Network Features

• The following are the benefits of RLC
  retransmission above the end-to-end reliability
  offered by TCP.
• The small PDU size used in RLC makes
  retransmission more efficient.
• The response time on the feedback from the
  receiver is smaller than is possible with TCPs
  end-to-end feedback.

8
Related Works (cont.)

• The locations of producing significant delay in
  WCDMA Networks
• RLC retransmission
• Processing delay for FEC and interleaving
• Buffer in RNC
• Problems for TCP in WCDMA Networks
• High bit error rates
• Delay-jitter

9
Related Works (cont.)

• Approaches for improving TCP performance over
  wireless mobile networks.
• Improvement of the existing TCP protocol
• Split connection approach
• Link layer solution

10
Related Works (cont.)

• According to prior research, we made the
  following assumptions.
• For optimizing TCP, only those techniques
  compatible with modern TCP implementations could
  be used.
• To preserve the end-to-end communication model,
  we do not adopt the intermediate node approach.
• A detailed model is needed that accounts for the
  use of selective repeat style ARQ.

11
Related Works

• Based on the above assumptions, we examined the
  link characteristics of the WCDMA network while
  setting the following goals.
• Suppress delay-jitter in link layer to avoid
  excess retransmissions
• Clarify the trade-off between jitter-suppression
  and link utilization to improve TCP throughput
• Optimize the link layer and TCP parameters

12
TCP performance with RLC retransmission (cont.)

• Bandwidth-delay product
• The optimal advertised receive window size is
  based on the product of the bandwidth and the
  delay of the link network.

13
TCP performance with RLC retransmission (cont.)

• Persistence in link layer retransmission
• Persistence in link layer retransmission involves
  a trade-off between IP packet loss and the
  efficiency of link utilization.
• MAX_DAT the maximum number of retransmission
  attempts for a single PDU

14
TCP performance with RLC retransmission

• Suppressing delay-jitter and increasing the
  available bandwidth
• RTO SRTT max(G, KRTTvar) where K 4
• Timer_Status_Prohibit (TSP) defines the delay
  imposed on the receiver before it can issue a
  STATUS PDU.

15
Performance Evaluation via Emulation and
Simulation (cont.)

• Evaluate TCP time out behavior of a
  representative operating system (OS) using the
  emulator.
• Evaluate the optimal TSP value from simulations.
• Change the maximum number of retransmission
  attempts (MAX_DAT) to change the link persistence
  to determine the relationship between TCP
  throughput and packet loss rate and thus obtain
  the optimal MAX_DAT.
• Determine the appropriate advertised receive
  window for TCP reflecting the WCDMA link BDP with
  simulated TCP traffic.

16
Performance Evaluation via Emulation and
Simulation (cont.)
17
Performance Evaluation via Emulation and
Simulation (cont.)

• The number of incidents of RTO vs BLER (using
  WCDMA emulator)

18
Performance Evaluation via Emulation and
Simulation (cont.)

• Throughput vs BLER (using WCDMA emulator)

19
Performance Evaluation via Emulation and
Simulation (cont.)

• Number of STATUS PDU and RTO vs
  Timer_Status_Prohibit (using Opent)

20
Performance Evaluation via Emulation and
Simulation (cont.)

• Throughput versus Timer_Status_Prohibit (using
  Opent)

21
Performance Evaluation via Emulation and
Simulation (cont.)

• Throughput and Down link segment loss rate vs
  MAX_DAT (using Opent)

22
Performance Evaluation via Emulation and
Simulation

• Throughput versus TCP Receive Window Size per
  BLER (using Opent)

23
Conclusion

• Linux is more aggressive to triggering
  retransmission in response to jitter than BSD or
  Solaris it allows RTO to occur more frequently.
• For the trade-off between this suppression and
  link utilization, the optimum TSP value is 200ms.
• PDU retransmission should attempted up to five
  times to hold the BLER 010 in simulated
  network.
• To performance of high latency networks, the TCP
  receive window should be set at 48 64 Kbytes.