SCOOP

SCOOP is a system monitoring tool for a PC Cluster System with the SCore Software Environment. The design concept of SCOOP is based on a multi-layered server-client model, which complements the scalability of the cluster.

These notes can be used by the administrator to use SCOOP on your PC Cluster System, and also to re-compile and install SCOOP.

SCOOP consists of three components: an agent program, a data collection server program, and a GUI tool for monitoring the data.

The agent program, scoopd(8), runs as a daemon on each compute host and collects system resource information from the Linux kernel.

The server program, ScoopServer(8), can run on a machine external to the cluster, or on one of the compute hosts, as long as the Java Runtime Environment is installed. ScoopServer communicates with any number of compute hosts running the scoopd agent, and periodically collects host resource information.

The GUI monitoring tool, Scoop(8), can run on a machine external to the cluster, or on one of the compute hosts, as long as the Java Runtime Environment with the JFC component is installed. Scoop communicates with any number of ScoopServer programs, and displays compute host activity as a block matrix, as well as compute host resource information as formatted table data. Scoop users can easily monitor activities of multiple clusters from a single viewpoint.

Here is a schematic diagram of the SCOOP system monitoring tool showing 32 compute hosts and one server host. ScoopServer, Scoop, scoreboard(8) and msgbserv(8) are shown all running on the same server host, although it is possible for each program to be running on different hosts:

[SCOOP schematic diagram]
PCCC logo PC Cluster Consotium
CREDIT
This document is a part of the SCore cluster system software developed at PC Cluster Consortium, Japan. Copyright (C) 2003 PC Cluster Consortium.