At 04:50 PM 9/6/1999 -0700, you wrote:
>The reason this is best is because the ADSM server will launch an
>additional server thread for each logical volume he has to manage.  And
>as we all know, the more threads a process has, the more parallelism
>employed.

Joshua,
Parallelism by itself may do nothing if all of the parallel threads are
going after the same shared resource (the disk drive).  In fact, if there
is logic in a physical disk volume manager thread to optimize I/O to the
disk (e.g., by minimizing seek times) then defining multiple LVs per pdisk
would actually be harmful.  The thread would be treating each LV as a
unique physical disk, which of course it is not.

One other thought-  if an ADSM thread is blocking while it waits for disk
I/O to complete, then I agree with your analysis: multiple LVs (and
therefore threads) would perform better.  Having multiple I/Os active to a
disk at once will yield higher disk utilization.

Without having access to the code, it seems to me that this could be
settled very easily with an empirical test.  The goal here is to maximize
utilization of a disk drive, yes?  That is easy to measure.  So, try it
both ways:  1 LV per PD, or multiple LVs per PD.  Keep other variables
constant, and measure.

A comment from the relevant ADSM server developer here would also settle
the issue.

..Paul