Strange output from DECps

[Kris Clippeleyr]

Hi folks,

A very strange phenonemon on 2 of our non-clustered DS20E systems.
Both systems are running V7.3-1, and have 512MB memory, and are running the same apps.
Both are recently AUTOGENed & rebooted. The MODPARAMS.DAT are identical apart from obvious things like SCS stuff.
Triggered by some complaints about performance, I started the Unicenter TNG Performance Advisor (formerly DECps) for both systems, and extracted the data of the past 24 hours. To my (and my colleague's) surprise, the graphs presented for memory usage were quite different for both systems.
On system SKG2A (although memory is used to the fullest), the graph made sense. On system SKG1A, it didn't, in that the portion taken up by the so-called "system working set" is approximately 50% of the memory.
We checked, and checked. Couldn't find significant differences in SYSGEN parameters, workload, or otherwise, apart from the fact that the SKG1A has 2 pagefiles, and the SKG2A only one.

Could any of you wizards and gurus, shed some light on this ?

I'm attaching some data regarding the systems (Excel spreadsheet with the data, SYSGEN parameters, and the like).

Any hints greatly appreciated, and rebooting the systems is no option (yet).

Thanx,

Kris

[Kris Clippeleyr]

Hi again,

The SYSGEN paramaters are in this attachment.

Greetz,

Kris.

[Ian Miller.]

whats the current size of pool on the two systems (SHOW MEM/POOL)?

[Kris Clippeleyr]

Ian,

Pools are on both systems:
nonpaged approx. 14.3 MB
paged approx. 5 MB
lock mgr approx. 21 MB

See attached file.

Kris

[Ian Miller.]

in the result for SHOW MEM on system is the last line
"Of the physical pages in use, 7785 pages are permanently allocated to OpenVMS."

The same?

[Kris Clippeleyr]

Ian,

On SKG1A:
Of the physical pages in use, 7785 pages are permanently allocated to OpenVMS.

On SKG2A:
Of the physical pages in use, 7802 pages are permanently allocated to OpenVMS.

Greetz,

Kris

[Ian Miller.]

can you compare SYS$SYSTEM:PARAMS.DAT on each system to see what differences. I'm just wondering if a difference has arisn parhaps as a result of autogen feedback or something.

[Wim Van den Wyngaert]

Kris,

Can't open your sysgen file (banking security).

What is the value of sysmwcnt on both systems ? Normally this value is autogened and for some reason too big.

Wim

[labadie_1]

can you do, on both systems
$ mc sysgen
set/out=node
sh/all
sh/spe
$ dif/para node1 node2
Apart from scsnode and scssystemid, do you have very differents numbers ?
Can you post the result of the diff?

[labadie_1]

The pql_dpgflquota and pql_mpgflquota are different, they should not

PQL_DPGFLQUOTA 200000 65536 -1 -1 Pagelets D
PQL_MPGFLQUOTA 200000 2048 -1 -1 Pagelets D
PQL_DPGFLQUOTA 65536 65536 -1 -1 Pagelets D
PQL_MPGFLQUOTA 32768 2048 -1 -1 Pagelets D

For Wim, sysmwcnt are similar
SYSMWCNT 6110 2048 512 1048576 Pagelets
SYSMWCNT 6322 2048 512 1048576 Pagelets

[Kris Clippeleyr]

Ian, Gerard, Wim,

All pertinent questions. The answer to them is already in my own reply to my original question.
A text file is attached to that entry, with the SYSGEN parameters of both systems.
Of course the settings differ somewhat, but not significant enough to explain why on the SKG1A system, DECps reports that the system working set occupies approx. 50% of the physical memory, and on SKG2A approx 10%.
As to the SYSMWCNT, it's larger on the "normal" system SKG2A (6322 pagelets) than on the SKG1A system (the one that's bugging me; 6110 pagelets).
The only noteable difference is the setting of the "special" system parameter S2_PAGES (10411 on no. 1, 4809 on no. 2).
Can this explain the difference reported by DECps?

Greetz,

Kris

[labadie_1]

the diff/para

ORION::$ dif/para 1.txt,2.txt
-----------------------------------------------------------------------------------------------------------------------------------
File DISK$ORIX:[SRC]1.TXT;1 | File DISK$ORIX:[SRC]2.TXT;1
-------------------------------- 2 --------------------------------------------------------------- 2 ------------------------------
MAXPROCESSCNT 287 32 12 163 | MAXPROCESSCNT 250 32 12 163
-------------------------------- 14 -------------------------------------------------------------- 14 -----------------------------
SYSMWCNT 6110 2048 512 10485 | SYSMWCNT 6322 2048 512 10485
internal value 382 128 32 655 | internal value 396 128 32 655
KSTACKPAGES 2 1 1 7 | KSTACKPAGES 2 1 1 7
BALSETCNT 285 30 8 163 | BALSETCNT 248 30 8 163
-------------------------------- 21 -------------------------------------------------------------- 21 -----------------------------
NPAGEVIR 59998208 8388608 163840 | NPAGEVIR 60129280 8388608 163840
PAGEDYN 5210112 524288 65536 | PAGEDYN 5390336 524288 65536
MPW_WRTCLUSTER 64 64 16 5 | MPW_WRTCLUSTER 64 64 16 5
MPW_HILIMIT 16384 512 64 | MPW_HILIMIT 16384 512 64
MPW_LOLIMIT 855 16 0 | MPW_LOLIMIT 744 16 0
-------------------------------- 32 -------------------------------------------------------------- 32 -----------------------------
FREELIM 305 32 16 | FREELIM 268 32 16
LAMAPREGS 0 0 0 2 | LAMAPREGS 0 0 0 2
LOCKIDTBL 36776 3840 1792 167769 | LOCKIDTBL 40360 3840 1792 167769
-------------------------------- 41 -------------------------------------------------------------- 41 -----------------------------
SCSSYSTEMID 1044 0 -1 | SCSSYSTEMID 1057 0 -1
SCSSYSTEMIDH 0 0 -1 | SCSSYSTEMIDH 0 0 -1
SCSNODE "SKG1A " " " " " "ZZZ | SCSNODE "SKG2A " " " " " "ZZZ
-------------------------------- 101 ------------------------------------------------------------- 101 ----------------------------
GH_EXEC_CODE 1536 1024 0 81 | GH_EXEC_CODE 1024 1024 0 81
GH_EXEC_DATA 312 512 0 20 | GH_EXEC_DATA 272 512 0 20
-------------------------------- 138 ------------------------------------------------------------- 138 ----------------------------
GBLPAGES 1203255 65536 10240 | GBLPAGES 1219752 65536 10240
internal value 75204 4096 640 | internal value 76235 4096 640
GBLPAGFIL 49408 512 32 | GBLPAGFIL 49408 512 32
QUANTUM 20 20 2 327 | QUANTUM 20 20 2 327
MPW_THRESH 1710 16 0 | MPW_THRESH 1488 16 0
-------------------------------- 169 ------------------------------------------------------------- 169 ----------------------------
FREEGOAL 1140 200 16 | FREEGOAL 992 200 16
GROWLIM 305 63 0 | GROWLIM 268 63 0
BORROWLIM 305 300 0 | BORROWLIM 268 300 0
-------------------------------- 236 ------------------------------------------------------------- 236 ----------------------------
PQL_DPGFLQUOTA 200000 65536 -1 | PQL_DPGFLQUOTA 65536 65536 -1
internal value 12500 4096 0 | internal value 4096 4096 0
PQL_MPGFLQUOTA 200000 2048 -1 | PQL_MPGFLQUOTA 32768 2048 -1
internal value 12500 128 128 | internal value 2048 128 128
-------------------------------- 260 ------------------------------------------------------------- 260 ----------------------------
ACP_MAPCACHE 178 9 2 | ACP_MAPCACHE 179 9 2
ACP_HDRCACHE 712 36 8 | ACP_HDRCACHE 862 36 8
ACP_DIRCACHE 712 22 4 | ACP_DIRCACHE 718 22 4
ACP_DINDXCACHE 178 26 2 | ACP_DINDXCACHE 179 26 2
-------------------------------- 268 ------------------------------------------------------------- 268 ----------------------------
ACP_QUOCACHE 287 64 0 23 | ACP_QUOCACHE 250 64 0 23
ACP_SYSACC 23 8 0 | ACP_SYSACC 20 8 0
-------------------------------- 339 ------------------------------------------------------------- 339 ----------------------------
LOCKIDTBL_MAX 489059 16776959 1792 167769 | LOCKIDTBL_MAX 493501 16776959 1792 167769
-------------------------------- 458 ------------------------------------------------------------- 458 ----------------------------

[Kris Clippeleyr]

Gerard,


The pql_dpgflquota and pql_mpgflquota are different, they should not

PQL_DPGFLQUOTA 200000 65536 -1 -1 Pagelets D
PQL_MPGFLQUOTA 200000 2048 -1 -1 Pagelets D
PQL_DPGFLQUOTA 65536 65536 -1 -1 Pagelets D
PQL_MPGFLQUOTA 32768 2048 -1 -1 Pagelets D


On another similar system (also DS20E, 512MB, VMS 7.3-1), the PQL_DPGFLQUOTA & PQL_MPGFLQUOTA are also set to 200000. This does not influence the size of the "system working set" as reported by DECps.
I have here 4 systems that I can compare, the SKG1A is the only one that "stands out" as far as DECps is concerned.

Greetz,

Kris

[labadie_1]

as nonpaged pool usage is different, check why with
$ ana/sys
sh pool/nonpaged/sum

Good hunt !

[Wim Van den Wyngaert]

May be related to this thread ?
http://groups.google.com/groups?hl=nl&lr=&ie=UTF-8&threadm=Qs-cndssgdWwRqKiXTWJkw%40comcast.com&rnum=3&prev=/groups%3Fnum%3D100%26hl%3Dnl%26lr%3D%26ie%3DUTF-8%26q%3Dopenvms%2B%2522permanently%2Ballocated%2522

Have a Duvel on me

Wim

[Ian Miller.]

Duvel is fine but I may prefer a Kwak :-)

I think the referenced cov thread is about excesive non-paged pool size which is not the case here - both systems have similar size pool. There are some differences in the GH_ params but nothing big.

According to a manual I've just looked SYSTEMWS is the % of physical memory used by processes with the username SYSTEM. So what processes run under username SYSTEM. Is there more of them on SKG1A or are they bigger and if so why?

[Kris Clippeleyr]

Ian,


According to a manual I've just looked SYSTEMWS is the % of physical memory used by processes with the username SYSTEM. So what processes run under username SYSTEM. Is there more of them on SKG1A or are they bigger and if so why?


Thanx for that tip.
I'll come back to this after the weekend.
Greetz, and have a Kwak on me.
I'm gonna indulge myself in a wee dram of single malt.

Kris

[labadie_1]

Are you talking about SYSMWCNT ? To see if it is too low, just

$ monitor page

let it some time (or all day), and check if peak system fault rate is different from 0. If it is, raise it and reboot.
System fault rate must be 0.

[Wim Van den Wyngaert]

Ian,

What manual are you talking about ? I'm at home but on my systems I found that the value reported by TNG were about the same as shown in "permanently" of show mem.

Wim

[Ian Miller. · Accepted solution]

I found a old hardcopy CA performance analyzer manual which SYSTEMWS was the % of physical memory used by processes with the username SYSTEM.

For the system being discussed the system working set figure on SKG1A is not the same as memory permenatly allocated - previously quoted as
on SKG1A:
Of the physical pages in use, 7785 pages are permanently allocated to OpenVMS.

On SKG2A:
Of the physical pages in use, 7802 pages are permanently allocated to OpenVMS.

[Kris Clippeleyr]

Ian,


I found a old hardcopy CA performance analyzer manual which SYSTEMWS was the % of physical memory used by processes with the username SYSTEM.


Via a SHOW SYSTEM, I found that on SKG1A 4613 pages out of 70943 pages are allocated to processes running under SYSTEM; on SKG2A, the figures are: 5311 pages out of 66768 pages.
These figures don't come near the values I got when displaying the memory utilization on SKG1A using DECps.
I'm very puzzled.

Kris

[Wim Van den Wyngaert]

Did the same test on a GS160 node running 7.3.

Used by SYSTEM processes : 141 MB.
Used by [1,1], [1,3] and [1,6] processes : 16 MB.

Reported by DECps : 126 MB

Reported by show mem (perm alloc) : 213 MB

Sysgen param SYSMWCNT : 6 MB

Puzzled puzzled puzzled

Wim

[Wim Van den Wyngaert]

Kris,

I stopped a process that took 70 MB (desta). In DECPS the SYSTEM WS decreased too. Permanently alloc however didn't change.

Perhaps DECPS is excluding global WS ?

Wim

[Kris Clippeleyr]

Guys,

I figured it out.
What happened.
The apps on SKG1A were started from the SYSTARTUP_VMS.COM, with a RUN/UIC=[100,1].
But since SYSTARTUP_VMS.COM is executed from a process with user SYSTEM, all apps run in processes with User: SYSTEM, & User identifier [100,1].
On SKG2A, the apps were restarted by an account PAM (uic [100,4]), although detached. So there, all apps run in processes with User: PAM, & User identifier [100,1].
So DECps is correct in saying that 50% or more of the physical memory is alloted to processes running as user SYSTEM on SKG1A, although NOT with UIC [1,4].

Greetz,

Kris

[Ian Miller.]

so the manual ment exactly what it said "processes running under USERNAME system" irrepective of UIC. A curios feature.