[Paper Review] Worst Case Buffer Requirements For Tcp Over ABR

Bobby Vandalore, Shivkumar Kalyanaraman+3 more|ArXiv.org|Sep 22, 1998

Network Traffic and Congestion Control1 references3 citations

TL;DR

This paper analyzes worst-case buffer requirements for TCP traffic over ATM's Available Bit Rate (ABR) service, focusing on scenarios where multiple TCP sources simultaneously send large bursts when congestion windows and allowed cell rates (ACRs) are high. Using analytical modeling and simulations, it demonstrates that the ERICA+ switch algorithm effectively controls queue lengths and buffer usage even under extreme conditions, ensuring stable performance in ABR networks.

ABSTRACT

ATM (asynchronous transfer mode) is the technology chosen for the Broadband Integrated Services Digital Network (B-ISDN). The ATM ABR (available bit rate) service can be used to transport ``best-effort'' traffic. In this paper, we extend our earlier work on the buffer requirements problem for TCP over ABR. Here, a worst case scenario is generated such that TCP sources send a burst of data at the time when the sources have large congestion windows and the ACRs (allowed cell rates) for ABR are high. We find that ABR using the ERICA+ switch algorithm can control the maximum queue lengths (hence the buffer requirements) even for the worst case. We present analytical arguments for the expected queue length and simulation results for different number of sources values and parameter values.

Motivation & Objective

• To evaluate buffer requirements for TCP traffic in ATM ABR networks under worst-case traffic conditions.
• To identify scenarios where large TCP congestion windows and high ACRs lead to excessive queue buildup.
• To assess the effectiveness of the ERICA+ switch algorithm in controlling queue lengths and buffer usage.
• To provide analytical and simulation-based validation of buffer stability under extreme load.

Proposed method

• Modeling TCP behavior under high congestion window and high ACR conditions to simulate worst-case traffic bursts.
• Extending prior work on buffer requirements with a formal analysis of queue length dynamics in ABR networks.
• Applying the ERICA+ rate-based congestion control algorithm at ABR switches to regulate cell transmission rates.
• Using analytical derivations to predict expected queue lengths under various source and parameter configurations.
• Conducting simulations with varying numbers of TCP sources and different ACR settings to validate theoretical findings.
• Comparing queue length performance across different parameter values to assess robustness of ERICA+ under stress.

Experimental results

Research questions

• RQ1 What is the maximum buffer requirement for TCP traffic over ABR under worst-case burst conditions?
• RQ2 How does the ERICA+ algorithm perform in controlling queue lengths when multiple TCP sources send bursts simultaneously?
• RQ3 What is the relationship between TCP congestion window size, ACR, and resulting queue buildup in ABR networks?
• RQ4 Can analytical models accurately predict worst-case queue lengths in ABR networks with TCP traffic?
• RQ5 How do varying numbers of sources and ACR values affect buffer stability under extreme load?

Key findings

• The worst-case scenario—large TCP congestion windows combined with high ACRs—results in significant queue buildup, but the ERICA+ algorithm effectively limits this growth.
• Analytical models accurately predict expected queue lengths under various source and parameter combinations.
• Simulations confirm that ERICA+ maintains bounded queue lengths even under extreme burst conditions.
• The buffer requirements remain manageable and predictable across different numbers of TCP sources and ACR settings.
• The ERICA+ algorithm demonstrates robustness in controlling queue lengths, ensuring stable performance in ABR networks.
• The study confirms that ABR with ERICA+ can support TCP traffic without excessive buffer demands, even in worst-case scenarios.