Have you checked the ADSM database cache hit ratio?  That is, in the
output of "query db f=d" this line...

                Cache Hit Pct.: 98.87


This number should be above 98%.  If not, then ADSM is spending a lot
of time seeking on disk for its internal database.

If you are under 98% then you need to increase the amount of memory
for the database cache.

Check out the performance tuning guide for how to do this:
        http://www.storage.ibm.com/software/adsm/pfguide/tgv31mst.htm#HDRBUFP
(the section on BUFPOOLSIZE)




And now the why [as I understand it]

The ADSM database resides on disk.  ADSM also caches commonly used
pages in memory for fast access and so it can do readaheads on the
database.


On platforms other than Novell, when the ADSM client starts a backup
it down loads from the ADSM server a list of all the files that were
present at the last backup.  This consumes a fair amount of memory on
the client and can be a problem if memory is tight.

Netware servers do not (until recently) have virtual memory so don't
cope with this well.

ADSM offers a "memory efficent backup" option that does not do that.
As it encounters directories it downloads only the files for that
directory.  This option is available on all platforms, and turned on
by default on Novell.

The effect of that is the Netware server is constantly making hits on
the ADSM database.  ADSM has to go back to physical disk to satisfy
these requests and this takes a long time.  (cf the download
everything approach where ADSM can use the internal cache to read
ahead of the clients and thus things run quickly).


I found that my Netware backups *crawled* along until I bumped up
BUFPOOLSIZE.  This literally started happening overnight as the load
on the ADSM server increased.  For a while it was fine, then one day
the Netware servers started crawling if more than one or two were
running concurrently.

The conventional wisdom in processor design is that a decrease in hit
rate of 1% gives a factor of 10 performance hit.  For ADSM it felt
like 1% gives a factor of 2.


Russell