1748283 Members
3743 Online
108761 Solutions
New Discussion юеВ

RMS buffer question

 
SAMI AHMAD
Regular Advisor

RMS buffer question

1)I set up the local rms buffer as "set rms/disk/system/buffer=20"

SUNNY2$ show rms
MULTI- | MULTIBUFFER COUNTS | NETWORK
BLOCK | Indexed Relative Sequential | BLOCK
COUNT | Disk Magtape Unit Record | COUNT
Process 0 | 0 0 0 0 0 | 0
System 32 | 0 0 20 0 0 | 8

Prolog Extend Quantity QUERY_LOCK
Process 0 0 System
System 0 0 Enabled

CONTENTION_POLICY
Process System
System Never


2)then i enable statistics for the file

set file sys$manager:SECURITY.AUDIT$JOURNAL;23/stat

3) if i monitor cache activity for the file I see nothing ..why ?

monitor rms/file=sys$manager:SECURITY.AUDIT$JOURNAL;23/item=cac


Active Streams: 0 CUR AVE MIN MAX

Local Cache Hit Percent 0.00 0.00 0.00 0.00

Local Cache Attempt Rate 0.00 0.00 0.00 0.00

Global Cache Hit Percent 0.00 0.00 0.00 0.00

Global Cache Attempt Rate 0.00 0.00 0.00 0.00

Global Buf Read I/O Rate 0.00 0.00 0.00 0.00

Global Buf Write I/O Rate 0.00 0.00 0.00 0.00

Local Buf Read I/O Rate 0.00 0.00 0.00 0.00

Cancel l Buf Write I/O Rate 0.00 0.00 0.00 0.00
7 REPLIES 7
Hein van den Heuvel
Honored Contributor

Re: RMS buffer question

What problem are you trying to solve, and what in this topic has anything to do with that?

I'm guess this is related to:
http://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=1319120

There is a suggestion, but no reason nor proof in there the auditing is part of a performance problem with heavy IO.

But why is this particular file picked on?

Anyway... if I wanted to see the read and write counts for that file, which btw as a performance expert on OpenVMS I never ever needed, then I would use two tools:

- $show device /file/out=show_device_system.log
- $search show_device_system.log audit
- $show memory /cach=(file=sys$manager:SECURITY.AUDIT$JOURNAL)

and

$ define PROCIO$SDA tools:PROCIO$SDA.EXE
$ anal/sys
OpenVMS system analyzer
SDA> set proc AUDIT_SERVER
SDA> procio

reads writes file - Process PCB: 883A5EC0
-------------------------------------------------------------
48 341 DISK$TD183SYS:[VMS$COMMON.SYSMGR]SECURITY.AUDIT$JOURNAL;1
8 0 DISK$TD183SYS:[VMS$COMMON.SYSMGR]VMS$AUDIT_SERVER.DAT;1
:



Back you your topic>>

1)I set up the local rms buffer as "set rms/disk/system/buffer=20"

WHY? How is that going to help your problem?


>> 2)then i enable statistics for the file
>> set file sys$manager:SECURITY.AUDIT$JOURNAL;23/stat

That would be for the CURRENT file and it should fail with a sharing violation.
Did you add /share?

For stats to be picked up, you need the process to re-open the file.
Did you use
$SET AUDIT/SERV=EXIT
$SET AUDIT/SERV=START

>> 3) if i monitor cache activity for the file I see nothing ..why ?

Maybe there is no activity?
Why do you believe this file is busy for you?

Hth,
Hein van den Heuvel ( at gmail do com )
HvdH Performance Consulting



SAMI AHMAD
Regular Advisor

Re: RMS buffer question

hi Hein!

thanks for your great info , I didnt use
$SET AUDIT/SERV=EXIT
$SET AUDIT/SERV=START

so i will try that now. I have one more question:
can I use rms buffers to improve IO performance of a non index file also ..not indexed ?
Robert Gezelter
Honored Contributor

Re: RMS buffer question

Sami,

Yes, the default RMS sequential file buffer sizes and counts are affected by the SET RMS command.

However, I must agree wholeheartedly with Hein's commnet: Nothing that has been cited in this (or the other) thread indicates that there is an IO bottleneck that will be fixed by changing the buffering factors on the Audit Journal.

While I have seen many situations where adjusting buffer factors and sizes had a significant impact, it is not a cure all. With all due respect, the accurate way to resolve a performance problem is to identify the actual bottleneck, and then take action to resolve it.

- Bob Gezelter, http://www.rlgsc.com
John Gillings
Honored Contributor

Re: RMS buffer question

SAMI,


>1)I set up the local rms buffer as
>"set rms/disk/system/buffer=20"

This is a very, very BAD idea. IMHO OpenVMS should prevent you from setting a non-zero value for system wide sequential disk buffers.

Here's why... this setting forces every seqeuential disk file opened by every process to be allocated 20 buffers. Blanket tuning "solutions" are rarely a good idea, but in this case there's a bigger problem.

For process permanent files, the buffers must also be process permanent. They must therefore be allocated from process permanent pool in P1 space. This is fairly limited in size (see SHOW PROCESS/MEM).

With a large enough value for sequential disk buffers, opening process permanent files (ie: OPEN from DCL) will deplete the pool and fail. It's not unusual for such a value to block batch jobs from running, because SYS$INPUT and SYS$OUTPUT are, by definition process permanent.

I strongly recommend you change the value to 0:

$ SET RMS/DISK/BUFFER=0/SYSTEM

If you really believe it's beneficial to have a non-zero value for all files, you can set a PROCESS value in LOGIN.COM or even SYLOGIN.COM.

$ SET RMS/DISK/BUFFER=20
A crucible of informative mistakes
SAMI AHMAD
Regular Advisor

Re: RMS buffer question

thanks everyone for your input , I got the idea . .stay away from tuning RMS buffers :)
Hein van den Heuvel
Honored Contributor

Re: RMS buffer question

Not really.

Tuning RMS buffers can be very rewarding... if there was an (IO) problem in the first place.

It is easy, may have great positive effect, and is low risk, but you have to have some clue of what is going on, or have someone to guide you.

Too many folks are clueless though. More than a few sequential file buffers is rarely useful, and may eat too much memory as John points out. But the default of 1 or 2 is too low. Try 4, not 20.... IF YOU HAVE AN IO PROBLEM AT ALL?!

Now for INDEXED files, the default depends on the file and is often 4 or 5, where 20 is often better. Global BUFFER may be better still!

For mere money myself, and others, will happily assist you in determining the basic performance characteristics of your application/system and an give you indication of what improvement can be made.
We'll be more than happy to teach you help you tweak RMS and other aspects of the box.

Cheers,
Hein van den Heuvel ( at gmail dot com )
HvdH Performance Consulting

Richard W Hunt
Valued Contributor

Re: RMS buffer question

The only I/O buffer tuning I ever set was for specific users - ORACLE for example, or a couple of my report writer products.

What I did instead was to check in SYLOGIN.COM whether this user was in a batch queue with lower priority than was allowed for interactive users. (GETJPI tells you most of what you need for that.) Then I did an "invisible" SET RMS for this session only. If interactive, no problem and you're on your own.
Sr. Systems Janitor