Biography
Izzy Grosof is an assistant professor at Northwestern University, in Industrial Engineering. They study stochastic queueing theory, with a focus on scheduling and resource allocation. Their research has received Best Paper awards from INFORMS, ACM SIGMETRICS and IFIP Performance.
Talk: “Scheduling with Massively Heterogeneous Resource Requirements”
Abstract: In modern computing systems, jobs resource requirements are massively heterogenous by default. Schedulers must allocate CPU, GPU, memory, and network resources, to name but a few. While some systems, such as cloud VM schedulers, allocate a handful of standardized resource bundles, many systems are flexible to exact user requirements, coping with the resulting heterogeneity. Existing queueing-theoretic performance analysis cannot handle this heterogeneity. I will present a family of scheduling policies designed to cope with practical resource heterogeneity, with proven optimal throughput and low simulated latency on datacenter trace data.