Hello. We have a couple of DEC 3000/400 Alphas running DEC OSF/1 V3.2
(Rev. 214) and configured with 64MB of memory. The local owners of these
systems are running some circuit-design programs on the systems, using
HSPICE Version 96.1 (960313) from Meta-Software,Inc., Campbell, CA.
The programs are taking roughly ten times longer than the users expected.
When the programs are running we have notice large amounts of disk activity
to the disks which contain the swap partitions. The virtual-address space
of the program is on the order of 100MB. Hence, we tentatively conclude
that the system is spending most of its time doing I/O to and from the swap
files.
On the other hand, it appears from the output of the "ps aux" command, for
instance, that not all of the available memory is being used.
The tech-support people at Meta-Software think that we must have the system
tuned incorrectly. Given that we pretty much left the kernel parameters as
they were when the system was installed, that's easy for us to believe.
Here's what appears to be the relevant section of the kernel-configuration
file:
#
# Special options (see configuring the kernel chapter
# in the Guide to System Administration)
#
timezone 0 dst 0
dfldsiz 134217728
maxdsiz 1073741824
dflssiz 2097152
maxssiz 33554432
cpu "DEC3000_500"
maxusers 16
machine alpha
I can find nobody around here who will admit to having any documentation
for the system, but I assume that the "*dsiz" parameters refer to what are
seen as the soft and hard C-shell limits for "datasize". Likewise, I
assume that the "*ssiz" parameters refer to what are seen as the soft and
hard C-shell limits for "stacksize".
I understand the maxdsiz to be the maximum virtual memory available to a
single process. Given that the number above is on the order of 1GB, that
would seem not to be a limitation here. I assume that maxssiz is that
maximum amount of system stack space available to a single process. Given
that its value here is on the order of 33.6MB, it does seem that this MIGHT
be a limiting factor for a process with 100MB address space.
Can anybody comment on this analysis? Thanks.
- Mike
----------
Michael Hannon hannon_at_physics.ucdavis.edu (Internet)
Dept. of Physics
University of California ucdhep::mike (42385::mike) (HEPnet)
Davis, CA 95616-8677 916-752-4966 (Voice) 916-752-4717 (FAX)
Received on Wed Jul 24 1996 - 01:59:35 NZST