As Agarwal describes the project by noting that "[d]atacenter transport design has usually been an algorithmic problem—designing congestion control protocols to achieve a multi-objective optimization goal (e.g., a combination of objectives like low latency, high throughput, etc.). This state of affairs is because, for the Internet and for early incarnations of datacenter networks, performance bottlenecks were primarily in the core of the network. Technology trends suggest that soon, if not already, this will no longer be the case." As a result, "adoption of Terabit Ethernet will fundamentally shift latency and throughput bottlenecks to the end-host."
Agawal notes that "[t]his project builds upon the observation that, when performance bottlenecks shift to end-hosts, datacenter transport design is not merely an algorithmic problem; instead, we need to take a more holistic approach: co-designing congestion control protocols with end-host mechanisms (hardware- driven admission and flow control, software and application schedulers) to efficiently utilize all resources." Consequently, his project will "explore such a holistic approach leading to an end-to-end design of datacenter transport for Terabit Ethernet."
Read related recent news about Agarwal, including: