Grid'5000 experiment

Jump to: navigation, search

Impact of number of computing nodes on time execution of MOTEUR. (Networking)

Conducted by

Tram Truong Huu

Description

Deploy the servers OAR controling 4, 8, 16, 32, 64, 128 computing nodes. Run MOTEUR( a workflow enactor) on these servers and computer the execution time. This work is composed of (i) deploy our environment on all of nodes by kadeploy, this step last out the long time (around 5 minutes for a node) if a node is not successfully deployed, we need to redeploy or reserve another node and (ii) execute workflow application "bronze standard" on MOTEUR for a growing size of the input data set (image pairs). The greatest makespan of this workflow is around an hour. However, it is not always successful due to the interaction of middleware components, between server and node computing, we need to redo sometimes.

Status

achieved

Resources

  • Nodes involved: 100
  • Sites involved: 3
  • Minimum walltime: >1d
  • Batch mode: no
  • Use kadeploy: yes
  • CPU bound: no
  • Memory bound: no
  • Storage bound: no
  • Network bound: no
  • Interlink bound: no

Tools used

kadeploy, MOTEUR

Results

Evaluation the makespan of application, the overhead of data transfers and the speed-up of application in function of the number of computing nodes.

More information here

Shared by: Tram Truong Huu
Last update: 2008-04-11 13:45:35
Experiment #448

Personal tools
Namespaces

Variants
Views
Actions
Public Portal
Users Portal
Admin portal
Wiki special pages
Toolbox