Grid'5000 experiment

Jump to: navigation, search

Benchmarking of network management plateforms (SNMP, JMX) (Networking)

Conducted by

Abdelkader Lahmadi

Description

Our work aims to assess the performance of management framworks on large scale environement such as grid 5000. In an earlier work, we have analysed the performance of JMX (Java management eXtension), the de facto standard for java application management, within a cluster (ICluster2 in grenoble) environement. Our main results shows that management delays follows a two parameters weibull distribution. We have also identified a scalability metric that puts in relation the quantity of management activities, the cost of these activities in terms of resources consumption and the quatlity of these activities. We have illustrated this metric to assess the scalability of a JMX based centralized management architecture. We extend our performance on the grid5000 environement to state the impact of a multi-sites agents architecture on the tested management framework.

Status

in progress

Resources

  • Nodes involved: >1000
  • Sites involved: >6
  • Minimum walltime: >1d
  • Batch mode: yes
  • CPU bound: yes
  • Memory bound: yes
  • Storage bound: no
  • Network bound: no
  • Interlink bound: no

Tools used

No information

Results

Not yet

More information here

Shared by: Abdelkader Lahmadi
Last update: 0000-00-00 00:00:00
Experiment #34

Personal tools
Namespaces

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