Thanks to all who replied so promptly, as always - this list is an awesome
resource.
I described a system with 2GB memory, 1999MB swap, asked if I should do
the work to give it more swap (all partitions are in use).
Mary Freesmeyer noted that we could add add'l swap partitions, allocating
spare partitions, and that it is possible to set the priority down on the
smaller add'l partitions so they are hit less often than primary swap.
Jay Leafey said:
[too little swap] can have a massive impact on a heavily-loaded system
and can prevent a lightly-loaded system from achieving it's potential
performance... The rule-of-thumb for swap space varies with the amount
of physical memory, tho, as sys_check wants 2.5 x physical memory on my
768 MB AlphaServer 4100.
(sys_check wanted 1 x the amount of avail. system memory after boot on our
2GB system)
Dr. Tom Blinn explained what memory would actually be needed, and suggested
leaving it alone for now:
If you are using "eager" swap, then you will never be able to use more
physical memory for user space (swappable) than the size of swap. So,
you might need larger swap. On the other hand, at least some memory is
NEVER available to users -- the kernel's instructions, stack, etc. and
any pages used for I/O buffers (file system space and so on). None of
this memory is paged. And, if you use "lazy" swap (not recommended on
systems where you are short on real memory and can't control the kind
of applications that are run), you can use all of physical memory PLUS
all of swap for pages (but again, not all of physical memory will be
available for user pages).
Unless you are really running out of swap, leave it alone. If you've
got the ability to add another disk to the system, just put swap on
two (or more) disks. Disks are cheap compared to the risk of breaking
something trying to repartition existing disks, and having more disk
spindles for swap improves performance.
In our case, we use eager swap so I believe we'd never actually run out,
thus I'm not going to do anything.
Thanks again!
--
Judith Reed
jreed_at_appliedtheory.com
(315) 453-2912 x5835
Received on Wed Feb 23 2000 - 15:16:13 NZDT