LoCal Cluster

From Local

Jump to: navigation, search

Contents

Workloads

  • Identify workloads: batch processing (Hadoop), stateless web services (web server/Cloudstone).
  • Characterize idle times
  • Identify "critical path" resources

Scheduler

  • As few nodes awake as possible, subject to SLA constraints
  • For the awake nodes, run as fast as possible
  • Schedule jobs to run concurrently if their resource demands are orthogonal

Hardware

  • Identify power efficient hardware
  • Characterize hardware power states and transition times

Software

  • Optimize software (both time and energy efficiency gains)
  • Configurations (lowest barrier)
  • Scheduling (low barrier)
  • Application architecture (more difficult)

Benchmarking and Analysis

  • Suspend resume times and costs

Presentations

  • Power-performance tradeoff (2009-10-21, LoCal meeting) slides
Retrieved from "http://local.cs.berkeley.edu/wiki/index.php/LoCal_Cluster"
Personal tools