Carnegie Mellon Parralel Computing Notes on Lecture 6

Another solution smarter scheduling schedule long

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: imality(x[i]); } Coarse granularity partitioning: 1 “task” = 10 elements Decreased synchronization cost (Critical section entered 10 times less) So... have we done better? CMU 15-418, Spring 2014 Rule of thumb ▪ Useful to have many more tasks* than processors (many small tasks enables good workload balance via dynamic assignment) - Motivates small granularity tasks ▪ But want as few tasks as possible to minimize overhead of managing the assignment - Motivates large granularity tasks ▪ Ideal granularity depends on many factors (Common theme in this course: must know your workload, and your machine) * I had to pick a term. Here I’m using “task” generally: it’s a piece of work, a sub-problem, etc. CMU 15-418, Spring 2014 Smarter task scheduling Consider dynamic scheduling via a shared work queue What happens if the system assigns these tasks to workers in left-to-right order? Cost 16 Tasks CMU 15-418, Spring 2014 Smarter task scheduling What happens if scheduler runs the long task last? Potential for load i...
View Full Document

This document was uploaded on 03/19/2014 for the course CMP 15-418 at Carnegie Mellon.

Ask a homework question - tutors are online