Abstract:
Modern cloud applications increasingly rely on low-latency communication, yet end-host bottlenecks remain a major barrier to achieving predictable performance. In this talk, we examine the problem of slow receivers at end-hosts, where limitations in CPU scheduling, networking stacks, and system interfaces can significantly degrade both latency and throughput in cloud VMs.
We first analyze how receiver-side inefficiencies manifest in real systems, particularly under high request rates and RPC-heavy workloads. We then present a set of techniques to mitigate these bottlenecks by rethinking end-host design, improving data movement, and enabling more efficient interaction between applications and the underlying system.
Our results demonstrate that addressing slow receivers at the end- host can lead to substantial improvements in both tail latency and overall throughput, highlighting the importance of end-host optimizations in modern cloud environments.
Speaker bio:
Alireza Sanaee is a systems software engineer and researcher working at the intersection of operating systems, networking, and memory systems. He received his PhD in Computer Science from Queen Mary University of London, where he was advised by Gianni Antichi at Queen Mary University of London and Brent Stephens at Utah/Google. His research focused on reducing end-host overheads for low-latency and RPC applications in data centers.
He previously worked as a researcher at National University of Singapore and Huawei Research in Cambridge, and is hosted at University of Cambridge under Prof. Andrew Moore.
His work spans the Linux kernel and QEMU, with an emphasis on low-latency system design and high-performance infrastructure. His research has been published in top systems conferences, including ASPLOS, ATC, NSDI.
Link to join online:
