Answers to my original post on the subject of 'page faulting'
------------------------------------------------------------------------
----------------------------------------
Tom Webster wrote:
================
vm-maxvas is the max amount of virtual memory that any non-kernel
process
can have at any point in time. You may want to try increasing this to
2GB. Don't forget to adjust vm-vpagemax, which should be the number of
8k pages (vm-maxvas/8192).
On a large system like yours, you may also need to change
vm-min-kernel-address and vm-kentry_zone_size. This generally needs to
be done only on systems with several gigs of memory and a lot of
processes.
Stephen Mullin wrote:
=================
maxVirtualAddressSpace set to > phy space .. i.e. close to phys + swap
ubc-max ... set to < ~20-25
SGA ... I/O buffers .. big as possible
advfs buff 1% ... yes
OTHERS:
# seqstart is % of phys mem NOT UBC
vm-ubcseqstartpercent = 25
# seqpercent is % of UBC size
vm-ubcseqpercent = 20
ubc-maxdirtywrites = 15
This also helps datafile adds and such, they (DEC) tell me.
AdvfsFavorBlockingQueue = 0
alan_at_nabeth.cxo.dec.com wrote:
==========================
A high rate of zero fill pages often indicates that you have a
program which is being started, having memory cleared for it
and then going off and doing its work. Continued high rate
of zero fill pages usually means that this program is being
started a lot. If it doesn't quickly fill the process table,
then they are exiting fairly quickly. So, a program that is
being started and quickly existing is a common cause. Look
through the various log files the system and applications
keep to see if it leaves any tracks. You might also be able
to use the system's accounting to track it.
------------------------------------------------------------------------
----------------------------------------
Thanks a lot guys.
Those were the answers to the following questions:
Dear sirs,
I have an as8400 with 4G ram running Oracle 7.3.3.6 with SGA 1GB
and I'm experiencing excesive page faulting. My sysconfigtab file has
the
following modifications:
proc:
maxusers=2048 ( provided by sys_check. Most of our connections are
through sqlNet)
vm:
vm-maxvas=1073741824 (1GB)
ubc-maxpercent=60
vm-ubcseqstartpercent=20
vm-mapentries=16384
vm-pagemax=65536
vfs:
bufcache=1
name-cache-hash-size=512
advfs:
advfsCacheMaxPercent=1
AdvfsMaxDevQLen=32
Note: There are very few (and small) page outs (1 in 100), many page ins
(100-300 pages)
and page faults that go up to 3000 pages. On vmstat output, the 'zero'
value is very close to
the 'fault', if that helps. System time upon high page faulting rises
abow 50.
Could anyone give a hint on how I could smooth this problem?
And another question. Since I'm running Oracle with an SGA of 1GB,
should I leave
the advfsCacheMaxPercent to 1 and let Oracle to it's cacheing?
(I use KZPSA to connect to HSZ50 wich controlles stripped RZ1CB-CS
disks)
____________________________
Alexander Sakellaris
Sr. Systems Administrator
Operations Department
IT Division
Network/Systems & Support Sector
COSMOTE S.A.
____________________________
Received on Tue Sep 01 1998 - 14:02:25 NZST