Event Information:

Date and time: July 16, 2019; 10:00-11:30
Location: Engineering 2 Room 215

Abstract:

While throughput can easily be measured, tail latency is harder to capture and characterize in a statistically meaningful manner, as it depends on a number of factors beyond the workload itself. In this talk, Marios will present LANCET, a self-correcting tool designed to measure the open-loop tail latency of μs-scale datacenter applications with high fan-in connection patterns. LANCET is self correcting as it relies on online statistical tests to determine situations in which tail latency cannot be accurately measured from a statistical perspective because of  the workload configuration, the client infrastructure, or the application itself. LANCET leverages hardware timestamping to measure RPC end-to-end latency, when available, in order to eliminate measuring errors coming from the client.

Bio:

Marios Kogias is a 4th year PhD student working with Ed Bugnion in EPFL. His main research focus is optimising for microsecond scale RPCs. He’s working on both building systems with strict tail-latency SLOs, and accurately measuring RPC latencies. He has interned at Cern, Google, and Microsoft Research, and he is an IBM PhD Fellow.