[Linux-cluster] journal size, performance

[Shawn Hood]

Hey folks,

I'm in the process of implementing a GFS cluster.  A quick over of our hardware:

1 Apple xraid (with plans to bring the two others into the SAN after testing)SA
3 Dell (2x 1950 / 1x 2950) boxes, running RHEL4u5
Qlogic SANbox
Qlogic HBAs

lvscan of SAN volume group

ACTIVE            '/dev/hq-san/nlp_qa' [700.00 GB] inherit
  ACTIVE            '/dev/hq-san/prod_reports' [1000.00 GB] inherit
  ACTIVE            '/dev/hq-san/cam_development' [500.00 GB] inherit
  ACTIVE            '/dev/hq-san/nlp_development' [500.00 GB] inherit
  ACTIVE            '/dev/hq-san/svn_users' [1.50 TB] inherit


We run a mixed environment of RHEL4.5 and RHEL5.1.  We're currently in
the process of developing a plan to shorten the OS lifecycle so we can
have all of our cluster nodes in a more homogenous RHEL5.1
environement.  Anyhow, we plan to scale to about 3-5x our current
storage capacity and 3-4x our current server count within the next
several months.

That said, I need to decide know the parameters to use when creating
journals--number of journals and size.  I could not find any
information about how the number of journals affects performance.
What kind of performance impact would I see by creating 9-12 journals
that we will eventually need?  If the size of the filesystem is
smaller than the logical volume, can the number of journals be
increased?  What about journal size impacting performance?

If information on this already exists, could someone point me to it?

Shawn