[an error occurred while processing this directive]

HP OpenVMS Systems

ask the wizard
Content starts here

RAID and I/O performance?

» close window

The Question is:

 
Hi,
 
We have a number of controller based RAID 5 sets configured via an HSZ50
ocntroller.
 
I have a question which has puzzled me and would be interested in any input
you have.
 
All things being equal, from purely a performance point of view, if we had 6
files all being updated at the same time by the same application, would be
it be better to to have a single logical VMS disk based on a 6 member RAID 5
set(using one disk from ea
ch of six available shelves) or would we benefit by splitting each file to
seperate partition on the RAID 5 set and mounting the six VMS logical disks
like this. Please assume that writeback and read cache are enabled on the
unit.
 
I get the feeling that it will boil down to a question of how the caches are
utilized, but lacking this information, I am unable to form an opinion.
 
Regards
 
Edwin
 


The Answer is :

 
  When a single application updates multiple files on individual partitions
  on a controller disk unit, it is effectivly guaranteeing a certain "seek"
  distance.  This is not optimal.  This does not involve the configuration:
  whether the unit was RAID-5, RAID-1, or JBOD, though with RAID at least
  there are some odds that the files will be distributed across different
  spindles.  In the opinion of the OpenVMS Wizard, the caching algorithms
  do not play a particular role here.
 
  When a single application needs to more or less simultaneously update
  multiple files, then either those files should be on independent disk
  spindles, or they should be located as close together as is possible.
  In the same partition, in other words.  You can further request the files
  be allocated nearby using the allocation attributes available via the FDL
  "POSITION" and associated subfunctions for "FILE" and LBN proximity.
 
  File placement is often the least effective of the available performance
  "tweaks".  Pre-allocation, correct bucketsizes, global buffers, faster
  I/O subsystems, and having the appropriate numbers of buffers available
  are all typically more effective controls.  But once all those settings
  are near optimal, then by all means consider the implications of file
  placement for that last measure of application performance.
 

answer written or last revised on ( 17-SEP-1999 )

» close window