Initiating various failures and tests

To start or stop the HACMP cluster service in a single node, you can issue the commands:

smit clstart

smit clstop

HACMP also provides a central administrative utility, so you can use the C-SPOC (Cluster Single Point of Control) utility to start the whole HACMP cluster by issuing the commands:

smit cl_clstart.dialog

smit cl_clstop.dialog

There are several places to look up log information: /usr/adm/cluster.log provides a high-level view of the current cluster status. It is a good place to look first when diagnosing a cluster problem. /tmp/hacmp.out is the primary source of information when investigating a problem.

You can also configure a remote machine to connect to the HACMP cluster and use clstat (or xclstat) to monitor and test the HACMP cluster.

We initiated various failures at all the possible failure points in our WebSphere test topologies.

For the HACMP cluster, we initiated the failures by:

Starting/stopping HACMP cluster services

Powering down the database server node

Disconnecting the network cable

Killing the database processes

Stopping the database servers

For the other parts in our WebSphere test topologies, we initiated failures by: Killing HTTP, application, administrative (Node Agent and Deployment Manager), and LDAP server processes Stopping HTTP, application, administrative (Node Agent and Deployment Manager), and LDAP server processes Disconnecting the network cable to each node Powering off each node one at a time

We initiated various kinds of failures in our test topologies, and we counted the failed client requests versus the total client requests. The failure rate was very small (<1%), depending on the frequency of initiation of various failures, HACMP or HACMP/ES detection and recovery time, and network characteristics. We found that WebSphere process failures were instantaneously recovered through the WebSphere WLM mechanism and contributed very little to failed requests. However, database recovery took minutes to complete and contributed the most part to the failed requests. Therefore, in the next section, we discuss how to tune the clustering parameters.

  Prev | Home | Next

 

WebSphere is a trademark of the IBM Corporation in the United States, other countries, or both.

 

IBM is a trademark of the IBM Corporation in the United States, other countries, or both.