Hi,
I guess you're talking abaout a database server doing asynchronous I/O or using raw devices?
Else such a setting would cost a bit too much filesystem cache, I think.
What are your "memory issues"?
Anyway, the command to use here is "vmo" (Virtual Memory Options).
To check your current settings issue
vmo -a ( vmo -F -a under AIX 6)
and examine maxclient%, maxperm% and minperm%.
To achieve your desired settings, issue
vmo -o minperm%=10
vmo -o maxperm%=15
vmo -o maxclient%=15
To make these settings permanent across reboots add the "-p" flag -
vmo -p -o minperm%=10
vmo -p -o maxperm%=15
vmo -p -o maxclient%=15
PERM memory is used for file pages, CLIENT memory is essentially the same (with some specifics for JFS2 and NFS, which can be ignored here). The remainder is in a way "reserved" for working storage pages (%Comp).
These settings will be applied immediately, but since AIX is somewhat "conservative" in its memory management it will need some time to adjust its memory usage.
Also, the values will not really be taken "seriously". If not needed for COMP, AIX will use more memory for PERM than is set with maxperm%.
To avoid this, and to speed up memory adjustment, you could additionally set strict_maxclient = 1 and strict_maxperm = 1.
I don't recommend this, because it will cause very high page in/page out activity for a while, and will withdraw all flexibility in memory management (with the strict settings there will never be more than maxperm% file pages, regardless of whether the rest is really needed for COMP or not).
wmp
