Search for: All records

The goal of this short document is to explain why recent developments in the Internet's infrastructure are problematic. As context, we note that the Internet was originally designed to provide a simple universal service - global end-to-end packet delivery - on which a wide variety of end-user applications could be built. The early Internet supported this packet-delivery service via an interconnected collection of commercial Internet Service Providers (ISPs) that we will refer to collectively as the public Internet. The Internet has fulfilled its packet-delivery mission far beyond all expectations and is now the dominant global communications infrastructure. By providing a level playing field on which new applications could be deployed, the Internet has enabled a degree of innovation that no one could have foreseen. To improve performance for some common applications, enhancements such as caching (as in content-delivery networks) have been gradually added to the Internet. The resulting performance improvements are so significant that such enhancements are now effectively necessary to meet current content delivery demands. Despite these tangible benefits, this document argues that the way these enhancements are currently deployed seriously undermines the sustainability of the public Internet and could lead to an Internet infrastructure that reaches fewer people and is largely concentrated among only a few large-scale providers. We wrote this document because we fear that these developments are now decidedly tipping the Internet's playing field towards those who can deploy these enhancements at massive scale, which in turn will limit the degree to which the future Internet can support unfettered innovation. This document begins by explaining our concerns but goes on to articulate how this unfortunate fate can be avoided. To provide more depth for those who seek it, we provide a separate addendum with further detail. 

The trend towards powerfully programmable network switching hardware has led to much discussion of the exciting new ways in which it can be used. In this paper, we take a step back, and examine how it should be used. 

Recent years have seen a slew of papers on datacenter congestion control mechanisms. In this editorial, we ask whether the bulk of this research is needed for the common case where congestion control involves hosts responding to simple congestion signals from the network and the performance goal is reducing some average measure of Flow Completion Time. We raise this question because we find that, out of all the possible variations one could make in congestion control algorithms, the most essential feature is the switch scheduling algorithm. More specifically, we find that congestion control mechanisms that use Shortest-Remaining-Processing-Time (SRPT) achieve superior performance as long as the rate-setting algorithm at the host is reasonable. We further find that while SRPT’s performance is quite robust to host behaviors, the performance of schemes that use scheduling algorithms like FIFO or Fair Queuing depend far more crucially on the rate-setting algorithm, and their performance is typically worse than what can be achieved with SRPT. Given these findings, we then ask whether it is practical to realize SRPT in switches without requiring custom hardware. We observe that approximate and deployable SRPT (ADS) designs exist, which leverage the small number of priority queues supported in almost all commodity switches, and require only software changes in the host and the switches. Our evaluations with one very simple ADS design shows that it can achieve performance close to true SRPT and is significantly better than FIFO. Thus, the answer to our basic question – whether the bulk of recent research on datacenter congestion control algorithms is needed for the common case – is no.