GPM system is a distributed monitoring system for clusters or Grid nodes. Construction of system is based on a hierarchical design.

Monitoring system is based on the concept of metric. Metric is the basic entity, that can be described by one value at any time. Metric's name is splitted into parts by symbol ':'.
There are some types of metrics:

Host metrics

These metrics descibe the state of host. Name of these metrics contains 4 logical sections:

host:<GridNode or Cluster Name>:<Host Name>:<Host metric name>

For example: 'host:UNN_Cluster:s-cw0-00:mem:avail'

Job metrics

These metrics descibe the job information. Name of these metrics contains 6 logical sections:

job:<Job Identifier (JOBID)>:<GridNode or Cluster Name>:<Host Name>:<Process identifier (PID)>:<Job metric name>

For example: 'job:12e28850-520d-11db-9ecc-c11185f8066c:UNN_Cluster:s-cw0-00:0000:mem:usage'

Other metrics

These metrics can describe other information (for example: from Grid services). The logical structure of this information is not specified, because this information is depended from many factors. We are planned to add the service's information.

The manager.properties file (directory is ./Distrib/manager) contains parameters for Manager

• gpm.manager.storage.dir specifies the storage folder. Directory must not contain other information.
• gpm.manager.port specifies the Manager port. This port is used by agents and different clients.
• gpm.manager.configuration.metric specifies the file name of XML file contains the configuration of stored metrics. See XSD schema file at schema/manager/MetricDirectory.xsd
• gpm.manager.neighbors=[manager01(server01:port01) | server02:port02] ...
  This parameter specifies the list of managers which store the same data. The data on these managers maintains in the equal state. Also configure Monitoring Agent's arguments with this list of managers addresses.

Run the script MonitoringManager.java.win.bat in the folder ./manager

Run the script ma.bat in the folder ./Distrib/ma_win or ./Distrib/ma_linux. The parameters '. "" ""' setup the external metric values producer. Second and third parameters are script names, which are used to obtain the values for external metrics. The first parameter is a working directory. The scripts are run in this directory. For additional info about scripts see files in then ./Distrib/ma_win/scripts or ./Distrib/ma_linux/scripts. subdirectories.

Run the gpm-monitoringclient.jar in the folder release. Also you can configure the Web Frontend to access via browser.

The GPM system contains advanced client part. System clients carry out the following tasks:

• Generation of requests to the Manager
• Reception of the information from the Manager
• Processing of the received information (a conclusion of tables, diagrams etc.)

The system has the following kinds of clients:

• Commandline Client. The elementary client realized as the command line utility. The user specifies a file which contains the text of request. The request is sent the Manager, and the answer received from it is deduced to the console. This Client can be useful at system debugging or reception of values of concrete metrics
• Monitoring Client. The most advanced client of GPM system realized as independent Java application. Monitoring Client uses socket mechanism for connection to the Manager, therefore it can be used from any computer of a local network. With Monitoring client the user can form complex requests on the basis of the list of available metrics. The data received from the Manager, can be represented as tables or diagrams. Some means of the analysis of the information are accessible also
• Applet Client. The client intended for working on a web page. Applet Client duplicates by opportunities Monitoring Client.

There are commandline tools that works with Manager and used to do:

• flush the Manager cache
• destroy metrics
• query the information
• receive notifications of request value's change
• check the current manager state (online)

The FlushManagerCache is a commandline utility that allows to stop Manager without lost the cached values.
Usage:

FlushManagerCache managerHost managerPort

The DestroyMetrics is a commandline utility that allows to destroy metrics. This operation is needed to free storage space and to destroy useless information about destroyed jobs a long time ago.
Usage:

DestroyMetrics managerHost managerPort metricsTemplate

The GetRequestValue is a commandline utility that allows to request information from Manager. You must specify the request file.
Usage:

GetRequestValue managerHost managerPort requestFile [parameters]

The SubscribeRequest is a commandline utility that allows to control the value of the specified request. Also you must specify the manager address and port.
Usage:

SubscribeRequest managerHost managerPort heartbeat requestFile [parameters]

The DumpManagerInfo is a commandline utility that allows to check current Manager state. You must specify the manager address and port.
Usage:

DumpManagerInfo managerHost managerPort

The project was started in Jan 2006. Today we have the first version of the system. The big work was done, but much remained unfinished. The nearest our plans:

• Develop Monitoring Agent for Linux
• Development of the web service for integration of several Managers
• Development of the client (or completion existing version) for working on multicluster level
• Integration of system with Globus MDS services

Copyright (C) 2006-2007 University of Nizhniy Novgorod

The GPM System Development Team:

Dmitry Labutin	Project Admin	labutin at users.sourceforge.net
Alexander Alekhin	Developer	alekhin at users.sourceforge.net
Denis Bogolepov	Developer	deniscmc at users.sourceforge.net