justin@bain3.oz (Justin Beech) (10/16/87)
What are the different ways (if there are different ways) that UNIX
allocates swap space for processes?
My experience (Pyramid OSx) is that the strategy is to allocate swap space
for the size of the text+data in the swap area UPON THE INITIATION OF THE
PROCESS. I assume shared text is only preallocated once.
This pre-allocated swap space can be a disaster for low disk to memory ratio
environments (eg 32MB of memory with 300MB of disk), especially if the
executables use sbrk lots, I am looking at a situation where we either
waste disgustingly large amounts of disk space, (space that will stay
virgin), or we watch sbrk fail on a machine with nothing swapped out,
and plenty of remaining free PHYSICAL memory: (ie maybe there are quite
a lot of processes active, but each is quite happy with a low percentage
of itself paged in).
Is this pre-allocation swap strategy outdated, common, the exception or what?
If this is outdated, what release of what Unix did it originate from,
or do these things vary according to how a port is done?
Maybe this is an old subject, or a trivial one, so its up to you more
experienced contributers as to whether to email me direct or reply
over the net.
------------------------------------------------------------------------------
Justin Beech ACS: justin@bain3.bain.OZ
Bain & Co Asset Financing, UUCP: {enea,hplabs,mcvax,uunet,ukc}!\
Sydney, NSW 2000 AUSTRALIA munnari!bain3.bain.OZ!justin