Ganglia Grid Monitor

1
Ganglia Grid Monitor
Federico D. Sacerdoti San Diego Supercomputer
Center University of California, San Diego
2
Ganglia Introduction

• What is ganglia?
• Efficient Cluster/Grid monitor.
• Why is it useful?
• Knowledge.
• My authority.
• I am a primary Ganglia developer.
• Monitor Core
• Web frontend
• Linux work

3
Ganglia Structure (gmon)

• Nodes Multicast metric values (UDP) within a
  cluster.
• Load_one, mem_free, cpu_sys, user-defined
• Each host has global knowledge of cluster.
• Efficiency
• Value thresholds
• Reliability
• Time thresholds
• XML querying
• Strong typing for viz
• Freshness stamps

Cluster nodes
Ganglia Multicast channel
4
Ganglia Structure (gmon)

• Nodes Multicast metric values (UDP) within a
  cluster.
• Load_one, mem_free, cpu_sys, user-defined
• Each host has global knowledge of cluster.
• Efficiency
• Value thresholds
• Reliability
• Time thresholds
• XML querying
• Strong typing for viz
• Freshness stamps

Cluster nodes
Ganglia Multicast channel
5
Meta Ganglia Structure (gmeta)

• Collects data from multiple clusters.
• Polls ganglia clusters using unicast (TCP).
• Redundant failover at cluster level.
• Every node has full cluster knowledge
• Keeps metric history.

Meta
A
C
B
Ganglia-enabled Clusters
6
Visualization

• Website parses XML from gmeta.
• Creates dynamic graphs on demand.
• Three historical views
• Grid view
• Cluster summary
• Host detail

Grid
Cluster
Host
7
Visualization
Physical view

• Physical view
• Shows node locations
• Grid tree view
• Displays Grid heirarchy

Grid Tree
8
Grid of Grids

• Enables arbitrarily-deep grid nesting in ganglia
• Distributes metric history collection, a proven
  bottleneck.
• Only keep metric history for child clusters.
• Web viz now a distributed system.

c
Grid A
Grid B
Meta
web
Meta
web
a
Meta
web
Grid C
9
Credits

• Primary Ganglia Developers
• Matt Massie (Berkeley)
• Founder, Core design, webfrontend design,
  Linux/Solaris ports.
• Federico Sacerdoti (SDSC)
• Webfrontend, core, grid-nesting, and metrics.
• Preston Smith (Purdue)
• Porting master FreeBSD, OS X, others
• Steve Wagner (ILM)
• Irix port, metrics, testing

• Mason Katz (SDSC)
• MDS provider
• Greg Bruno (SDSC)
• Ganglia python client
• And many others
• http//ganglia.sourceforge.net