There has been much discussion recently about ADSM throughput.  I
was wondering what definition of throughput was being used and
how is it being calculated.  The reason I ask is that we are
getting rather poor end to end throughput, about 12 kbyes/sec.
The definition I am using is based on the data in the SMF records
(session duration and kbytes transmitted).  Andy Raibeck has
pointed out that this measure is problematic because ADSM
probably spends more time scanning drives for changed data than
actually backing them up.  Indeed, we are experiencing on the
average about 70% of each session spent in a wait state.
Nonetheless, if one wanted to estimate how long it would take to
back up a certain amount of data, wouldn't the end to end
throughput be the best basis for making such an estimate?  Others
on this list have commented on the large proportion of session
time spent waiting.  We would certainly be very interested in
getting our 70% waiting time figure down.  Does anyone have any
data on the impact of changing the client's tcp/ip buffer size on
the proportion of time spent waiting?  If most of the waiting
time is due to scanning the disk, I don't see how tuning can help
much.

------------------------------------------------------------------
Robert P. Klein                          KL2 AT CU.NIH DOT GOV
Robert P. Klein                          KL2 AT CU.NIH DOT GOV
Phone: 301-496-7400                      Fax: 301-496-6905
Mail:  DCRT/CFB/ETS, 12A/1033, 12 SOUTH DR MSC 5607,
       BETHESDA, MD 20892-5607
------------------------------------------------------------------
=========================================================================