insufficient dynamic memory

[Lokesh_2]

Hi,

While submitting a batch job , I am getting the following error. I increased JTQUOTA, PFQUO for the user but with no success. Ran AUTOGEN too, but that also has not given any relevant warning. Now I do not know what to do...except reboot ..

%SUBMIT-F-CREJOB, error creating job
-SYSTEM-F-INSFMEM, insufficient dynamic memory

Thanks & regards,
Lokesh

[Ian Miller.]

Check non-paged pool - is there some free and how fragmented is it.

[Lokesh_2]

Hi Ian,

This is a three node cluster. Here is the output requested

__________________________________________
System Memory Resources on 29-JAN-2004 12:56:42.61

Nonpaged Dynamic Memory (Lists + Variable)
Current Size (bytes) 22241792 Current Total Size (pages) 43441
Initial Size (NPAGEDYN) 22241792 Initial Size (pages) 43441
Maximum Size (NPAGEVIR) 98850816 Maximum Size (pages) 193068
Free Space (bytes) 12662208 Space in Use (bytes) 9579584
Size of Largest Block 1429504 Size of Smallest Block 64
Number of Free Blocks 57801 Free Blocks LEQU 64 Bytes 173
Free Blocks on Lookasides 57099 Lookaside Space (bytes) 9111616

Paged Dynamic Memory
Current Size (PAGEDYN) 11999744 Current Total Size (pages) 23437
Free Space (bytes) 9604704 Space in Use (bytes) 2395040
Size of Largest Block 9578960 Size of Smallest Block 16
Number of Free Blocks 310 Free Blocks LEQU 64 Bytes 286


System Memory Resources on 29-JAN-2004 12:56:48.46

Nonpaged Dynamic Memory (Lists + Variable)
Current Size (bytes) 21649408 Current Total Size (pages) 42284
Initial Size (NPAGEDYN) 21600256 Initial Size (pages) 42188
Maximum Size (NPAGEVIR) 96000000 Maximum Size (pages) 187500
Free Space (bytes) 12648256 Space in Use (bytes) 9001152
Size of Largest Block 1469312 Size of Smallest Block 64
Number of Free Blocks 24369 Free Blocks LEQU 64 Bytes 80
Free Blocks on Lookasides 23501 Lookaside Space (bytes) 3777024

Paged Dynamic Memory
Current Size (PAGEDYN) 11999744 Current Total Size (pages) 23437
Free Space (bytes) 9530736 Space in Use (bytes) 2469008
Size of Largest Block 9491744 Size of Smallest Block 16
Number of Free Blocks 282 Free Blocks LEQU 64 Bytes 263


System Memory Resources on 29-JAN-2004 12:56:54.28

Nonpaged Dynamic Memory (Lists + Variable)
Current Size (bytes) 21600256 Current Total Size (pages) 42188
Initial Size (NPAGEDYN) 21600256 Initial Size (pages) 42188
Maximum Size (NPAGEVIR) 96000000 Maximum Size (pages) 187500
Free Space (bytes) 10871744 Space in Use (bytes) 10728512
Size of Largest Block 6676416 Size of Smallest Block 64
Number of Free Blocks 18748 Free Blocks LEQU 64 Bytes 296
Free Blocks on Lookasides 17824 Lookaside Space (bytes) 2776640

Paged Dynamic Memory
Current Size (PAGEDYN) 11999744 Current Total Size (pages) 23437
Free Space (bytes) 9620864 Space in Use (bytes) 2378880
Size of Largest Block 9609440 Size of Smallest Block 16
Number of Free Blocks 303 Free Blocks LEQU 64 Bytes 274


_______________________________

Thanks & regards,
Lokesh

[Brad McCusker]

NPAGEDYN may need to be increased.

Check CTLPAGES and CTLIMGLIM too. CTLPAGES affects pool, and CTLIMGLIM affects amount of pool for image structures.

[Antoniov.]

Hi Lockesh,
some questions are borning in my mind.

1.st Have you problem in interactive mode or only in batch mode?

2.st Have your user enough memory?
You could view account information
$ ACC/USER=/SINCE=/FULL
Look at Peak Page File, Peak Working Set and Page Fault. It's not easy but you could evaluate memory for every user.

3.th You declare ran autogen; do you see into report (SYS$SYSTEM:*AGEN*.*)? What you see into it?

Remember virtual memory affect virtual memory system parameter, paged memory system parameter, page file quota of user and workink set; you have a lot variable to check!

Bye
Antoniov

[labadie_1]

Hello

You sould check Ctlpages for the process queue_manager is enough.

As you can't do
$ sh proc/memory/id=xxx

you can use
1) the Dcl version

DCL for OpenVMS - Example C - How To Use SHOW PROCESS/MEMORY For Another Process


http://h18000.www1.hp.com/support/asktima/operating_systems/CY-1027624800-1.html

2) the Sda extension ctlpa available at

DCL for OpenVMS - Example C - How To Use SHOW PROCESS/MEMORY For Another Process

http://h18000.www1.hp.com/support/asktima/operating_systems/CY-1027624800-1.html

[Lokesh_2]

Hi Antoniov,

Here are the answers:

1. This user is only used in non-interactive mode. actually a scheduler (polycenter) job is running everyday which submits about 20 odd batch jobs with this username. Till yesterday, this was working fine, from today onwards giving problem. I tried to run this job on two nodes of the cluster, both given same error.

2.accounting I have not used yet..

3 I ran autogen on all the three nodes of cluster , and only one node has suggested to increase npagedyn. I have increased that and will reboot the cluster now.

Apart from npagedyn, I have increased CTLPAGES too. Also, increased jtquota & pgflquo in uaf .

I hope, a reboot of the cluster will solve the problem.

Thanks & regards,
Lokesh

[Lokesh_2]

Hi Labadie,

My systems are VAX & also do not have C compiler. The link provided by you is for alpha only.

Best regards,
Lokesh

[Ian Miller.]

I think one of the links shold have been this
http://h18000.www1.hp.com/support/asktima/operating_systems/CY-1021490401-1.html

Which contains a DCL based solution

[labadie_1]

No Ian, the Dcl contains a line

$ if f$getsyi("ARCH_NAME") .nes. "Alpha" then exit

You can of course comment it, but it will not work :-(
Anyway, I think you should raise Ctlpages (and may be Piopages) and reboot.

[Antoniov. · Accepted solution]

Hi Lokesh,
what does happen after yesterday?
Have you exhausted virtual memory (it seems very difficult)?
I think you have some problem with queue manager. You could try a brutly way (if you can stop system):
1) list the queues on system (with all details using SHO QUE /FULL);
2) stop all queues (STOP/QUEUE/NEXT que>);
3) stop queue manager (STOP/QUE/MAN/CLUST);
4) restart new queue manager (START/QUE/MAN/NEW/ON:)
5) init all queues.

Use this way only if you don't find any other solution because you need a fwe time to stop and init queues; with /NEW you create a new queue manager with a new file so if old file is too fragmented queues start better and quicker than now.

Bye
Antoniov

[Mac Lilley]

Hello

Does scheduler create detached processes or batch processes? If it creates detached processes you should check the values of PQL_D parameters

$ mc sysgen sho/pql

you may need to raise the value of PQL_DJTQUOTA (it's dynamic)


HTH

ML

[labadie_1]

This applies to vax and Alpha


[OpenVMS] SYSTEM-F-INSFMEM Error From DCL PRINT, SUBMIT and Queue Commands

http://h18000.www1.hp.com/support/asktima/operating_systems/0094EC6F-7778AF60-1C0097.html

and this is interesting

[OpenVMS] What Does the INSFMEM Condition Return from $SNDJBC Mean ?


http://h18000.www1.hp.com/support/asktima/operating_systems/009454D9-F73D4040-1C0181.html

[Lokesh_2]

Hi Everybody ,

Problem resolved !! ( without reboot ) . I guess Antoniov clue did the trick. On checking my queues on the system , I found that one queue was stopped and was holding 1000's of jobs in it. I have deleted all the jobs in the queue, then deleted the queue, and then recreated the queue. Also stop and restarted the queue manager.

After doing this my scheduler job is successfully submitting jobs again.

Thanks to all for your kind suggestions,
Best regards,
Lokesh

[labadie_1]

One sidenote about the Dcl

http://h18000.www1.hp.com/support/asktima/operating_systems/CY-1021490401-1.html

It works fine on Alpha Vms. It fails on Itanium Vms because

repeat/until=NOTINPHYS"

is never encountered, as it is another error message!

I discovered this at the HP Forum for developpers, where I gained an Itanium.

Porting to Itanium Vms can be very easy (Wasd port was nearly immediate, see on http://wasd.vsm.com.au/) but you can have some differences.

[Ian Miller.]

RE DCL - but can it work on a VAX?

[labadie_1]

> can it work on Vax

I am afraid no.

I am sure about one thing: the line

exam ctl$gq_allocreg
is good on Alpha (gQ for Quadword)
but has no meaning on Vax

exam ctl$gl_allocreg
(gL for longword) can be tried, but I think it will not work.

regards

gerard