Hi,
I have been looking at a 2100 machine that runs Progress.
This machine is a dual processor 275MHz 2100, with 704MB of memory, and
three KZPAAs each with 6 or 7 drives on each chain :-)
SCSI 2 has RZ16, RZ17, RZ18, RZ19, RZ20, RZ21, RZ22, all RZ28s or 28Ms
SCSI 3 has RZ24, RZ25, RZ26, RZ27, RZ28, and RZ30, all RZ29s
RZ24 thru RZ28 is one 20+GB LSM volume :-)
SCSI 4 has RZ32, RZ33, RZ34, RZ35, RZ37, RZ38, the first 4 are RZ28(M)s, rest
RZ29s
AdvFS is used all over the place for the fast recovery after a crash,
although crashes are few and far between. LSM is used for the drives RZ24
thru RZ28.
The database is a production database that sustains both update and query
activity during the day. The system runs about 18 hours a day, and at this
time of the year, 7 days a week.
Using PYPM and the displays I can see the following:
All of memory is used, but there is little paging. Some times paging spikes
up to 600-800 pages per second, but real page ins and outs are very low.
UBC_MAXPERCENT is set to 60 (down from 100), so I conclude that memory is
not really a problem, but that the buffer cache grabs all the free memory it
can, and the system has to spend time getting it back. I could probably
reduce UBC_MAXPERCENT to 40 or 30.
CPU does not seem to be a problem, the CPUs seem to have 30-40% idle time
most of the time.
However, SCSI-2 gets enormous amounts of IO, with PYPM showing BLK IO xfers
per second of 50 or more for many of the drives all the time. I estimate
that the xfers per second for the buss is constantly over 200.
SCSI-3 is usually only carrying a load of around 50 xfers per second, as the
LSM set is used for on-line backup, so during the night, the databases are
backed up (disk-to-disk) to the 20+GB of space, then the LSM area is backed
up to tape.
This means that the load on SCSI 3 is usually caused by two drives, the one
that the backup is currently working on, and the drive at the end that seems
to sustain about 25 xfers per second.
SCSI-4 seems to sustain a load of around 150 xfers per second or less.
My feeling at this stage is that the way to squeeze more out of the machine
is to move two of the drives on SCSI-2 down to SCSI-3 while at the same time
moving two of the
However, in the longer term, as the databases continue to grow, and the load
on the system continues to grow, I think that I should recommend the following:
1. An upgrade to a 2100A is out. The extra PCI controllers will not give you
any more SCSI IO as you would not put additional SCSI controllers on the
last 4 PCI slots.
2. More memory on the machine does not seem to promise much as well.
3. Moving to KZPSAs and HSZ50s may improve things, however, the customer seems
wary of RAID ...
4. A 4100 with multi-channel ultrawide SCSI controllers and new disks is going
to cost a bundle.
Can anyone comment on using databases on HSZ40s or HSZ40s? Can anyone
comment on the above?
Regards
-------
Richard Sharpe, sharpe_at_ns.aus.com
NS Computer Software and Services P/L,
Ph: +61-8-8281-0063, FAX: +61-8-8250-2080, WWW:
http://www.ns.aus.com
NS is an OS-neutral, equal-HTTP-server opportunity employer!
Received on Tue Apr 22 1997 - 06:18:05 NZST