Reducing the TCP Acknowledgment Frequency

By: 
Sara Landström and Lars-Åke Larzon
Appears in: 
CCR July 2007

Delayed acknowledgments were introduced to conserve network and host resources. Further reduction of the acknowledgment frequency can be motivated in the same way. However, reducing the dependency on frequent acknowledgments in TCP is difficult because acknowledgments support reliable delivery, loss recovery, clock out new segments, and serve as input when determining an appropriate sending rate. Our results show that in scenarios where there are no obvious advantages of reducing the acknowledgment frequency, performance can be maintained although fewer acknowledgments are sent. Hence, there is a potential for reducing the acknowledgment frequency more than is done through delayed acknowledgments today. Advancements in TCP loss recovery is one of the key reasons that the dependence on frequent acknowledgments has decreased. We propose and evaluate an end-to-end solution, where four acknowledgments per send window are sent. The sender compensates for the reduced acknowledgment frequency using a form of Appropriate Byte Counting. The proposal also includes a modification of fast loss recovery to avoid frequent timeouts.

Public Review By: 
Dina Katabi

They acknowledge every other segment, reducing overhead for both the end hosts and the network. This paper investigates whether one can further reduce ack traffic without degrading performance. TCP acknowledgments have multiple roles: they support reliability, trigger the transmission of new segments, and help in determining an appropriate send rate. The paper analyzes how ack frequency impacts the sending pattern and interacts with these roles. Interestingly, it shows that, in theory, TCP can achieve full utilization with just two acks per window at the cost of using significantly larger buffers. Using simulation, the paper shows that sending four acks per window produces good performance and is more robust in practice. The reviewers were concerned that TCP ack overhead is negligible. Thus, reducing the number of acks seems to solve a non-existent problem. Some reviewers felt that the ideas are useful in wireless networks, where link quality may be asymmetric and competing for medium access can create a significant overhead, even when the packets themselves are small. Despite these concerns, the reviewers felt that the ideas in the paper are intriguing, particularly the observation that TCP can make do with only two acks per window. Though reducing ack traffic may be a theoretical exercise, the paper provides enough insight to warrant publication.