I asked about the accuracy of what "collect -s pm" is reporting for swap
utilization (T64 5.1a, PK2), as man page says the swap figure is total
free, but on our system which doesn't appear to be swapping at all, that
value is sitting at 67 (MB?) when there is 8GB total swap.
I rec'd several responses, thanks to all.
Fred N. van Kempen speculates that collect may be reporting pure
overhead space used (data structures) - the actual amount of disk space
used (in blocks), not the number of pages swapped
Dr. Tom Blinn advises:
* check vmstat "pout" output on first line - total # of pages swapped
since last reboot - if this number is low, you have not swapped many
pages.
* check "swapon -s" output for available space - "If you have lazy swap
as the policy, then you may have almost all your swap available and none
of it allocated (since the only time pages get allocated with lazy swap
will be when you actually have to swap things out, and even then, the
pages get freed if the process that owns them terminates)" and further
notes "I really have to run the system out of physical memory to get it
to swap, since most of the time the VM subsystem will free up the clean
pages (and write back the dirty pages) from the universal buffer cache
(UBC) and free up enough memory, and also discard clean pages from
process' address spaces that are mapped to the file system (e.g., code
pages from parts of the programs that are no longer needed in their
active working sets since they are no longer executing in that part of
the code)."
Alan Rollow says
"67 MB is 8,576 pages, which is a value that doesn't show
up in the swapon listing. If the value is what is claiming
to be then it should be consistent with something seen from
swapon -s. Looks to me like Collect has a bug."
and finally, Jesper Frank Nemholt (HP) says:
"It's a bug in Collect.If you upgrade to the latest patchkit you'll get
a collect that reports correctly."
Received on Wed Oct 08 2003 - 13:00:52 NZDT