zemon@felix.UUCP (Art Zemon) (10/08/87)
Ultrix still has the old and now thoroughly outdated algorithm of allocating swap space to a process when the process is spawned instead of when the process actually needs to swap. For some environments this is absolutely absurd -- and expensive. For instance, we have an 8700 with 32 Mb of physical memory, 64 Mb of swap space that we never use because we never swap and almost never page, and we need MORE swap space because processes are dieing for lack of core (brk(2) fails)!! Does anyone know if anyone inside DEC, or even a third party vendor, is working on an updated memory/swap space management scheme to eliminate this problem? I really don't want to buy more storage which will never be used. This algorithm made sense when people used PDP-11s with 256K of memory and many megabytes of disk and they always swapped at least some. Now it is cheaper to buy enough memory that I never swap but I still have to buy the darned disks! Ahhhhh..., that feels better. Maybe I can still an old IBM-PC 30 Mb disk on the 8700 and call it swap space. :-) -- Art Zemon By Computer: ...!hplabs!felix!zemon By Air: Archer N33565 By Golly: moderator of comp.unix.ultrix
jdn@mas1.UUCP (Jeff Nisewanger) (10/16/87)
Reply-Path: The folks at Berkeley are rewriting the virtual memory code for BSD 4.4 for this and other reasons, so maybe it will be done in 2-3 years. I wouldn't look soon for a third-party fix. The memory management code in Ultrix (4.3 BSD) is not very modular and would be hard for an outsider to replace. Jeff Nisewanger Measurex Automation Systems pyramid!voder!mas1!jdn
mouse@uunet.UU.NET (der Mouse) (11/18/87)
Reply-Path: In article <8550@felix.UUCP>, zemon@felix.UUCP (Art Zemon) writes: > [about Ultrix using the usual UNIX swap space allocater] > For instance, we have an 8700 with 32 Mb of physical memory, 64 Mb of > swap space that we never use because we never swap and almost never > page, and we need MORE swap space because processes are dieing for > lack of core (brk(2) fails)!! How sure are you that brk() is failing for lack of swap space? There are three or four other reasons why it could lose. I suspect the likeliest is that you are running into the process size limit. To check this, try eating up swap space by running other things and see if you run out of memory at the same point. If you are hitting the size limit, try "ulimit datasize". If that doesn't help, you may have to rebuild your kernel with a larger limit. I could help you do this on 4.3, but Ultrix? Dunno, sorry. der Mouse (mouse@mcgill-vision.uucp)
zemon (Art Zemon) (12/08/87)
In article <13289@felix.UUCP> mouse@uunet.UU.NET (der Mouse) writes: > >How sure are you that brk() is failing for lack of swap space? I know it is brk(2) because of the messages coming from the failing programs. Furthermore, "pstat -s", when it can get enough core to run, reveals that we are critically low on swap space. -- -- Art Zemon By Computer: ...!hplabs!felix!zemon By Air: Archer N33565 By Golly: moderator of comp.unix.ultrix