Intel® Cluster Checker 3 runs slowly over NFS

In some instances, Intel® Cluster Checker 3 may take a long time to execute.

This can be caused by accessing the Intel® Cluster Checker database file on an NFS filesystem. A significant reduction in database transfer rate over NFS has been confirmed. On larger clusters, total execution time may be hours or days. By comparison, execution time for the same database on a locally-mounted filesystem would be seconds or minutes.

By default, Intel® Cluster Checker will store the database file, "clck.db", in the user's /home directory. Most clusters, including all Intel® Cluster Ready-compliant clusters, use a shared home directory. The shared /home directory is often mounted as an NFS share on each node. If /home is a local filesystem on the node where Intel® Cluster Checker is run, this article should not apply.

Not all NFS implementations experience this issue and Intel is working to determine a root cause.

To work around this issue, do not run Intel® Cluster Checker using a database file located on an NFS share. The database file should be relocated using one of the following methods:

  • Set the <database_file> element in the clckd.xml and clck.xml configuration files.
  • For analysis only, use the clck-analyze command switch "-D" or "--db" to specify the database file. There is no equivalent command line switch for clck-collect.

If using a database file on an NFS share, first copy the file to a local filesystem before using it.

Для получения подробной информации о возможностях оптимизации компилятора обратитесь к нашему Уведомлению об оптимизации.
Возможность комментирования русскоязычного контента была отключена. Узнать подробнее.