This is on our Oracle unix server---memory use is not really the problem, per se.(16 processors, 9216MB of real memory, 6144MB of paging space).
The problem is that computational pages are being paged out/in based on what we calculate as computational pages at about 50%, and not (no where near either) on the maxperm% setting of 20% (minperm% is at 10%).
Is this not correct?? Computational pages should NOT be paged out? If the working set of computational pages as a % of real memory is less than the MINPERM setting, then computational pages ARE fair game for stealing (paging out) to replenish the free list. But in our case they are greater than MAXPERM they should not be stolen. I've done daily svmon -G throughout the day and see computational as about 50%. Now, the very day we upgraded from 5.1 to 5.2 this started happening---whereas (under 5.1) this was not the behavior.
I don't believe setting strict_maxperm to 1 is what we want to do?? We believe that we would either waste memory (maxperm setting) if set too low, or cause computational pages to be paged out if set too high. In theory if strict_maxperm is not set, the file pages (numperm) should sort of dynamically adjust based on the requirements set for %computational pages?
Basicly how do we get back to the 5.1 behavior without actually downgrading to 5.1? Does anyone know what parm changes might make this happen?
Many thanks to any help you can give me!
The problem is that computational pages are being paged out/in based on what we calculate as computational pages at about 50%, and not (no where near either) on the maxperm% setting of 20% (minperm% is at 10%).
Is this not correct?? Computational pages should NOT be paged out? If the working set of computational pages as a % of real memory is less than the MINPERM setting, then computational pages ARE fair game for stealing (paging out) to replenish the free list. But in our case they are greater than MAXPERM they should not be stolen. I've done daily svmon -G throughout the day and see computational as about 50%. Now, the very day we upgraded from 5.1 to 5.2 this started happening---whereas (under 5.1) this was not the behavior.
I don't believe setting strict_maxperm to 1 is what we want to do?? We believe that we would either waste memory (maxperm setting) if set too low, or cause computational pages to be paged out if set too high. In theory if strict_maxperm is not set, the file pages (numperm) should sort of dynamically adjust based on the requirements set for %computational pages?
Basicly how do we get back to the 5.1 behavior without actually downgrading to 5.1? Does anyone know what parm changes might make this happen?
Many thanks to any help you can give me!