dolson@ADA20.ISI.EDU.UUCP (03/03/87)
Hi folks. I'll describe a situation I noticed over a year ago but since
until recently it hasn't caused any grief I haven't been too worried.
Briefly, since we converted from VMS V3.7 to V4.2, and upon subsequent
upgrades (soon 4.5), our page count in MODIFIED PAGES has been miniscule.
My understanding was that when pages were forced out of a working set
by process constraints, they were placed on the modified page list just
in case they might be needed again soon. Our 11/785s have 6M and 10M,
and memory constraints have only just begun to make themselves felt.
Under V3.x, we usually had several hundred pages on the modded page list.
Now, its seldom over 10. Thats what bothers me.
Several thoughts/interpretations present themselves:
1) VMS V4 AUTOGEN uses a different paging model and times out the
pages on the modded page list sooner than it did in V3.
2) workload changes due to our substantial remodeling during the
conversion cause smaller working sets, and pages never get
pushed out due to process WSMAX exceeded. (? this would be flaky
and I don't really think this is it.)
3) We broke something in the way we arrange memory (is 2500 SRPS
a problem? DBMS really chews 'em up) and it shows up as described.
(I obviously don't know the significant differences between paged
and non-paged memory). The systems are clustered.
Right now, we don't have a problem (uh, I hope). But medium term
requirements will dictate Oracle and All-in-1, so I want to prepare
for more memory consumption (yes, 32M are on order for each machine,
for you folks about to tell me about all-in-1) and I want to know how
to handle it when it gets here. Thoughts and speculation from all
welcomed.
Doug Olson
AFCMD/SCF
Kirtland AFB, NM
-------