Messages in this thread | | | From | Jesse Pollard <> | Subject | Re: Virtual vs. physical swap & shared memory forks (clone) | Date | Sat, 25 Mar 2000 20:00:27 -0600 |
| |
On Sat, 25 Mar 2000, Rik van Riel wrote: >On Sat, 25 Mar 2000, Linda Walsh wrote: > >> The idea is *predictability*. Guarantees of behavior. >> Your user deamon is fine for many cases, but it's execution is >> not deterministic. > >Please back up your assertions with code. If you can implement >a non-overcommit option which doesn't put overhead in the normal >kernel, I'm sure people will use it. > >The IRIX vswap argument isn't a really valid one either. >Almost all IRIX admins I've talked to had their vswap at >either zero or infinity because otherwise behaviour would >be too difficult to predict ... exactly the opposite of >what you are saying here.
That is only when they really don't understand the consequences.
I run systems both with vswap at 0, where I don't want any failures - disk servers, and at infinity - single user workstations where it doesn't matter, AND at something in between where I can (or am directed to) tolerate an ocasional failure.
The vswap operation was put back right after IRIX implemented an uncontroled overcommitment of memory resources. It took an entire version of IRIX to get it to work right, but as of now, it is very usefull. SGI lost several customers during the interim, but I think they started getting them back after the controls were working (we dropped a power challenge array after about a year and a half due to various problems that could have been caused by poor resource control).
------------------------------------------------------------------------- Jesse I Pollard, II Email: pollard@cats-chateau.net
Any opinions expressed are solely my own.
- 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/
| |