Hello,
I'm a little late on this thread but I feel that I have some practical
experience that may be relevant to the discussion. Among other tasks, I
manage a pool of a hundred or so dedicated compute servers. These are
HP-UX workstations, with 768Mb or more ram and @2-3Gb total virtual memory.
The compute farm is very stable even though systems are routinely overloaded.
This is because the program which tries to exceed swap+memory is the one
which gets back the NULL pointer and nothing is randomly killed. The
behavior is deterministic, making the boxes easy to maintain.
I doubt I'd trade that behavior for a model in which the process which
overcommits succeeds nominally, but then some process is randomly killed
later.
If fairness, the character of the workload contributes to the stability,
where they are "large" relative to system processes like sendmail, bind,
or inetd. We tend to overcommit by a large amount rather than a few pages.
Dram and disks are getting progressively cheaper, meanwhile availability
remains a precious commodity. I'd be much happier with a conservative
vm allocator than an aggressive one.
skd
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Thu Mar 23 2000 - 21:00:37 EST