doelz@urz.unibas.ch (Reinhard Doelz) (10/05/88)
*Problem*: On an IRIS 4D/80 with 370 megs hard disk and 12 megs of memory, the swap space is 50 megs, and the command swap -l displays that there are usually over 80% free. Big applications in the batch tend to crash, dupreg - insufficient memory to allocate ... pages, however, these are running nicely online. *Question*: How to make a job swap on the disk if others request more memory? I can't imagine that only the physical memory of the core may be used in the batch. Any comments/suggestions/flames welcome. Reinhard ************************************************************************ * Dr. Reinhard Doelz * SWITZERLAND * * Biocomputing * * * Biozentrum * doelz%urz.unibas.ch@relay.cs.net * * Klingelbergstrasse 70 * * * CH-4056 Basel * * ************************************************************************
yohn@tumult.SGI.COM (Mike Thompson) (10/13/88)
. *Problem*: . On an IRIS 4D/80 with 370 megs hard disk and 12 megs of memory, the . swap space is 50 megs, and the command swap -l displays that there are . usually over 80% free. Big applications in the batch tend to crash, . dupreg - insufficient memory to allocate ... pages, however, these . are running nicely online. . . *Question*: . How to make a job swap on the disk if others request more memory? I can't . imagine that only the physical memory of the core may be used in the batch. . . Any comments/suggestions/flames welcome. . . Reinhard When a System V process is exec'd, it preallocates the AMOUNT of swap space it would need if all swappable pages did, indeed, have to be swapped, but it does not allocate ACTUAL swap blocks. So "/etc/swap -l" indicates only how many blocks currently contain swapped-out blocks. The ``/debug'' line of a "/bin/df" will show you how many blocks (512 bytes) have been preallocated (amount, not actual blocks) and how many have not. While this may be a distinction without a difference, the applications aren't ``crashing''; exec() system calls are failing (errno EAGAIN) for lack of swap space. The process does have the opportunity at this point to make an orderly retreat. Mike Thompson SGI