> [mailto:[EMAIL PROTECTED]]On Behalf Of
> Jonathan F. Dill
> Sent: Tuesday, July 13, 1999 5:20 PM
>
> AFAIK unless you've done something to the kernel to get around that
> limit. What's the point of running swap on RAID anyway? Memory is
> cheap these days--seems to me rather than wasting time boosting swap
> performance, it would be a lot more efficient to get some more memory,
> unless you're running quantum mechanics simulations and are
> maxed out at 512 MB on your motherboard or something.
There are a number of reasons to do this. Most of them involve the
operation of a server in a production environment and having to maintain
a reasonable availability level (90%+).
You are correct in that, for performance reasons, production hosts are
tuned to never have to hit the swapper. Usually this is done by throwing
RAM at the box. However, should you become resource starved, for any
reason, the swapper is available for over-flow. RDBMS's are notorious
for high instantaneous memory peaks, as an example. The swapper allows
for a graceful degradation by handeling the over-flow. This is
preferable to running out of RAM (ugly).
However, as has been noted before, if any disk housing a swap partition
dies then the kernel crashes.Ergo, you toss the swapper unto a RAID1
partition and eliminate that SPOF... instant up-time improvement. Yes,
RAID1 is a little slower, but performance isn't the issue since, if you
are hitting the swapper with any frequency at all, on production
equipment, you should be looking at how to add RAM to the box.
