Paging stats - comments please

[Ranga Nathan]

Here is some info on paging for our production Linux environment. It
looks OK to me. However with 5G to go I wonder why there is any paging
at all! I am sure someone here will spot something here.
indicate pag all
17:03:28 MAINT 000000:000000 ZLT3 000000:012057 ZLT1
000000:008163
17:03:28 ZLP3 000000:007888 ZLP2 000000:013196 ZLP1
000000:009832
17:03:28 DATAMOVE 000000:000416 DIRMAINT 000000:001567 VSWCTRL1
000000:002582
17:03:28 TCPIP 000000:002660 VMSERVR 000000:001178 VMSERVU
000000:001178
17:03:28 VMSERVS 000000:001323 OPERSYMP 000000:001260 EREP
000000:001230
17:03:28 ZLT2 000000:020308
Ready; T=0.01/0.01 17:03:28
q stor
17:04:21 STORAGE = 5056M

We cycled the z/VM environment about 8 days ago when we did a POR.

ZLP1, ZLP2 and ZLP3 are our production environments. These machines are
sized 512M, 256M and 128M respectively. There is no CPU cap on any of
them. All test environments ZLT1 etc are 128M.

--
__________________
Ranga Nathan
Work: 714-442-7591

[Rob van der Heij]

While IND PAG ALL is not very helpful in analyzing your Linux
performance, it does suggest that you have chosen to ignore the
recommendation to configure expanded storage (or failed to limit MDC
in expanded storage).

Many things beyond virtual machine primary address space will compete
for residence in real memory. In many situations it is helpful that CP
will page out portions of the virtual machine that you do not use.
Having pages out on paging space does not impact performance unless CP
picked the wrong pages. If CP does pick the wrong pages, it helps to
have them go through expanded storage first.

When part of the virtual machine was paged out to disk and is
referenced again later, it will be paged in by CP. As long as the
virtual machine only reads the page and does not modify it (most of
your executables and shared libraries) that virtual machine resides
both in memory and on paging DASD.

Whether you're are paging and to know if that is impacting your Linux
performance needs a performance monitor, especially if you want to
increase utilization of your VM system.

Rob
--
Rob van der Heij
Velocity Software, Inc

[Eginhard Jaeger]

Well, if you don't have a performance monitor but want to know about paging
activity for a specific virtual machine then issue 'INDICATE USER userid' a
few times. This will tell you about the number of pages that are currently
used, and about the number of pages read and written in the meantime. And on
systems with some expanded storage it will also tell you about the machine's
use of expanded storage (you probably should define some).

While 'INDICATE PAG ALL' will tell you about pages allocated on DASD, you
really worry about those only when they frequently have to be brought back
into main storage and cause page waits. To see whether that is a problem
issue the 'INDICATE QUEUES' command repeatedly: it will tell you whether
your Linux guests suffer from frequent page waits that are slowing them
down.
Note, however, that statistically relevant wait state information is best
obtained from CP monitor data. And the easiest way to convert CP monitor
data to numbers you can understand is to use a performance monitor ..

Eginhard

----- Original Message -----
From: "Ranga Nathan" <***@cox.net>
To: <VMESA-***@LISTSERV.UARK.EDU>
Sent: Tuesday, February 21, 2006 1:14 AM
Subject: Paging stats - comments please

Here is some info on paging for our production Linux environment. It looks
OK to me. However with 5G to go I wonder why there is any paging at all! I
am sure someone here will spot something here.
indicate pag all
17:03:28 MAINT 000000:000000 ZLT3 000000:012057 ZLT1
000000:008163
17:03:28 ZLP3 000000:007888 ZLP2 000000:013196 ZLP1
000000:009832
17:03:28 DATAMOVE 000000:000416 DIRMAINT 000000:001567 VSWCTRL1
000000:002582
17:03:28 TCPIP 000000:002660 VMSERVR 000000:001178 VMSERVU
000000:001178
17:03:28 VMSERVS 000000:001323 OPERSYMP 000000:001260 EREP
000000:001230
17:03:28 ZLT2 000000:020308
Ready; T=0.01/0.01 17:03:28
q stor
17:04:21 STORAGE = 5056M
We cycled the z/VM environment about 8 days ago when we did a POR.
ZLP1, ZLP2 and ZLP3 are our production environments. These machines are
sized 512M, 256M and 128M respectively. There is no CPU cap on any of
them. All test environments ZLT1 etc are 128M.
--
__________________
Ranga Nathan
Work: 714-442-7591