I have a machine DEC alpha au (640 MHz with 1 GB RAM) running OSF1.
I mainly use the machine with programs which perform diagonalization
of big complex matrices using dynamical memory allocation
and LAPACK (dxml) routines.
The problems start when the matrix dimension reaches 4000x4000
(I have already changed the parameter MAXSSIZ in the kernel file
bringing it to 800000000 (initally it was set to 33554432)). However,
even after recompiling the kernel (via the command 'doconfig')
the problem was not solved. While the job is running, the maximum
memory used does not exceed the 750MB (I checked those parameters
through the VSZ and RSS values obtained via the command 'ps v'),
but the program stops with the message 'segmentation fault' while
executing one of the standard lapack routines. The same program
runs succesfully if the matrix size is 3000x3000.
At the moment the hard limits set for the kernel are as following:
>>limit -h
cputime unlimited
filesize unlimited
datasize 1048576 kbytes
stacksize 781250 kbytes
coredumpsize unlimited
memoryuse 1021648 kbytes
vmemoryuse 1048576 kbytes
descriptors 4096
I also tried to change the sysconfigtab file. Initially the last part of the
file was
set (by default, I believe) in the following way:
ipc:
shm-mni = 1024
shm-seg = 256
shm-max = 67108864
vm:
vm-mapentries = 1000000
proc:
************************************************
Now I changed it to:
ipc:
shm-max=1073741824
shm-mni=1024
shm-seg=1024
ssm-enable-core-dump = 0
vm:
ubc-minpercent = 10
ubc-maxpercent = 70
vm-vpagemax = 524288
vm-mapentries = 32000
vm-page-free-reserved = 4000
vm-page-free-swap=2065
vm-maxvas = 1073741824
proc:
maxusers = 64
max-proc-per-user = 64
per-proc-stack-size = 1073741824
per-proc-data-size = 1073741824
per-proc-address-space = 1073741824
max-per-proc-stack-size = 1073741824
max-per-proc-data-size = 1073741824
max-per-proc-address-space = 1073741824
However, the problem remains.
In addition, I can say that if I run the program using static memory allocation
it runs correctly also for matrices 4000x4000; this makes me believe that
the problem must be related to some parameter governing the dynamical usage
of the memory.
I would be very much appreciate any help
and/or suggestions you could give me in order to solve the problem.
Thank you so much for your help
Alessandra Continenza
--
Alessandra Continenza
Dip. Fisica - Univ. dell'Aquila
Via Vetoio 10, I-67100 L'Aquila (ITALY)
tel: (+39)-0862-433055 - fax: (+39)-0862-433033
Received on Tue Mar 07 2000 - 17:36:24 NZDT