2016 Poster Sessions : Core-Aware Scheduling: Balancing Application Concurrency with Core Availability

Student Name : Henry Qin
Advisor : John K. Ousterhout
Research Areas: Computer Systems
Abstract:
Kernel thread multiplexing is too slow for applications with latency requirements in the single microseconds. Polling is a poor workaround for the problems of kernel threading because it burns up cycles and bottlenecks application throughput. User threading suffers from priority inversions and core loss during blocking system calls. We propose core-aware scheduling, in which the kernel allocates cores to applications and a user-level library schedules user threads across these cores. This work is still very nascent but micro-benchmarks look promising.

Bio:
Henry Qin is a 4th year Phd candidate advised by John Ousterhout. Prior to the current work on core-aware scheduling, he worked on understanding and improving latency for RAMCloud, a distributed, low-latency key-value store originally built by his predecessors. He has also interned three times since starting his Phd. When Henry is procrastinating, he likes to answer questions on and participate in the StackOverflow community. Henry received his B.A. in computer science from Duke University in 2011.