HP 3000 Manuals

High End System Performance Considerations [ COMMUNICATOR 3000/XL XL RELEASE 3.0 ] MPE/iX Communicators


COMMUNICATOR 3000/XL XL RELEASE 3.0

High End System Performance Considerations 

by Tim Canepa 
Commercial Systems Division 

As processor speed increases so does the relative load placed on the
system.  There are a few items to keep in mind when configuring your new
high end HP 3000/980 system.  Since the IO bandwidth on the newer
machines is basically unchanged, you may want to consider a few simple
guidelines when adding memory and disk storage to your system.

The working set for a particular application will probably occupy a
somewhat static portion of main memory.  The user working set, however,
will occupy space based on the number of active processes running the
application.  If your memory does not scale sufficiently to support the
new user working set load, an unusually high level of swapping activity
may result, causing a decrease in overall system performance.  For
example, if the user working set occupies .6MB per user and the
application working set is 30MB, a 200 user system will require
approximately 150MB, not counting the overhead required by the system.
If you jump to 400 users, the working set space will occupy approximately
270MB (.6 x 400 + 30).  Attempting to run the 400 users in 192MB will
probably result in sluggish response times.  By scaling the memory size
appropriately, you will avoid unnecessary disk IO caused by swapping
transient space to disk.

Mixing disk sizes should be carefully monitored to achieve higher
performance levels.  If different size disks are used within the same
volume set, your performance will probably improve if the master volume
is one of the smaller disks (unless it is a 7935, which is relatively
slow).

The master disk of a volume set has a higher volume of disk traffic due
to the number and type of system structures that reside on it (the
transaction management logs reside on the master volume of a volume set).
Because the general disk space management policy allocates extents on the
disk with the most free space, larger disks tend to fill up faster
(percentage wise) than smaller disks in the same volume set.  This causes
the larger disks to incur more IO traffic than the smaller ones.

Since your master volume already has a high level of IO traffic, using a
larger disk as the master will probably decrease your overall performance
if the rest of the disks in the set are smaller.  This situation can be
compensated through the addition of memory, depending on the application
working set.  If you have a volume set that mixes disks with sizes of
1.3GB (5239568 sectors) and 544MB (2232192 sectors) you can expect the
distribution of space to be somewhat unbalanced.  The larger sized disks
have 3007376 more sectors of capacity thus will receive the major portion
of the extent allocations as well as the majority of IO traffic.  If the
larger disks are on a faster link (fiber link verses HP-IB) than the
smaller disks, the increased IO activity may be overcome by the high
performance link.  A general rule would be to make larger capacity disks
members, rather than masters if the disks in the volume set are using the
same type of link.



MPE/iX Communicators