Folks, during the small MDC we had a week ago, I was able to make
some useful measurements of HPSS data rates:
1) the peak data transfer rate of 500 Mbyte files from opus to HPSS
is between 15 and 20 Mbytes/sec. This is the rate at which
we can fill the HPSS disk cache. This rate is the ultimate limit
of our ability to record data.
2) the maximum sustained data rate into HPSS is around 8 Mbytes/sec.
At this rate the HPSS data migration to tape is just about keeping up
with data arriving from opus.
This rate is limited by the number of available
tape drives (one) and performance of the single drives (~10 Mbytes/sec
per unit).
The sending scripts have been tuned to use the feed-back information
on available HPSS disk cache space- if available HPSS disk space
drops below 10% of the total, file movement from opus to HPSS
stops until more space becomes avaiable.
3) the "RCF/HPSS design data rate" for BRAHMS is 4 Mbytes/sec. The sending
scripts have been tuned to send an average of 4 Mbytes/sec of data
(send one 500 Mbytes file at the 20 Mbytes/sec peak rate, then
sleep N seconds to bring the average rate down) and the system
was operated in this mode for 4 days (Wednesday to Sunday).
During this time there was only one 1 hour interruption in the data
transfers caused by a STAR tape drive failure.
4) RCF have developed a very useful data transfer monitoring tool,
at http://ru10.rcf.bnl.gov/alarms/daq.html
Conclusions: I am happy with the performance of the system. We are
able to move data between opus and HPSS at decent speed
(15-20 Mbytes/sec out of 20 Mbytes/sec maximum for the opus disk array),
HPSS is able to move data to tape at close to tape speed (8 Mbytes/sec
out of 10 Mbytes/sec tape limit) and the system appears to be rock
solid at design data rate (4 Mbytes/sec).
--
Konstantin Olchanski
Physics Department, Brookhaven National Laboratory, Long Island, New York
olchansk@bnl.gov
This archive was generated by hypermail 2b29 : Sun Mar 19 2000 - 23:25:30 EST