Palirria: accurate on‐line parallelism estimation for adaptive work‐stealing

Abstract

We present Palirria, a self-adapting work-stealing scheduling method for nested fork/join parallelism that can be used to estimate the number of utilizable workers and self-adapt accordingly. The estimation mechanism is optimized for accuracy, minimizing the requested resources without degrading performance. We implemented Palirria for both the Linux and Barrelfish operating systems and evaluated it on two platforms: a 48-core NUMA multiprocessor and a simulated 32-core system. Compared to state-of-the-art, we observed higher accuracy in estimating resource requirements. This leads to improved resource utilization and performance on par or better to executing with fixed resource allotments.

A self‐adjusting task granularity mechanism for the Java lifeline‐based global load balancer library on many‐core clusters

“…All the approaches mentioned above focus on spreading the work to keep as many computing resources busy for as long as possible. In Palirria, 18 Varisteas and Brorsson tackle the reverse problem consisting of matching the computing resources to the (varying) degree of parallelism of an application. On single applications, they are able to maintain ideal performance with higher efficiency by adjusting the number of allocated cores to the computation based on its potential for parallelization.…”