How do I increase free list size

[Paul cotton]

A picture speaks a thousand words, even if it is a picture of words....

Please attach a .TXT file with the output for $SHOW MEM/FULL when the system was nicely used, SYSGEN SHOW /MAJOR and toss in a SHOW SYSTEM or MONI SYSTEM (per process memory) output while collecting.

It is quiet possible there simply is not enough physical memory.

And focus on history.
You suggest is worked fine unattended. Well, stop attending!

What else changed? More users?

Were the working set params tweaked such that now memory is overcommited?
Was a dababase cache tweaked, making it use more memory?

Hein


I should point out that previously this system only hosted rdb databases but was recently upgraded to support oracle and rdb server (+ apparently system performance
was never that good in the first place)

I think the physical memory constraint on this box is, as you suggested, preventing me from increasing the free page list. I have however increased the modified page list on this system as I believe that if free list size becomes exhausted it can aquire unused pages from the modified page list ?

regards
Paul.


The attached file contains the output you requested.

[Volker Halle]

Paul,

please do not repeat the previous text in your replies. This wastes a lot of space on the screen, if looking at the whole topic.

The memory data shown in your attachment shows 347795 free pages = 2.8 GB (note: ALPHA 8kb pages !) of 4 GB of physical memory.

The pagefile space is not used at all, but it seems rather small (325 MB).

What do you do if the system hangs ? Just restart it ? Maybe consider to force a crash:

press halt button
>>> crash

Then you can look at the dump to figure out, what's really wrong...

Volker.

[Hein van den Heuvel · Accepted solution]

>> The attached file contains the output you requested.

It's a fine dry-run.
But the the system was entirely idle.
I indicated "when the system was nicely used".
Maybe I should have written 'badly used' instead of 'nicely' :-).

It does show something big already though.
It shows 1GB set aside for the new Oracle install.
On 4 GB that would rate as 'significant impact'
"reserved Memory" is the optimal way to use memory... if you have enough.
But is totally, extremely, inflexible.
It is set aside for one purpose and one purpose only and firmly suggest you know better then OpenVMS how to utilize memory.
The very fact that you opened this topic suggest that there is not enough knowledge to make that call.

Tell those Oracle folks to back off!
Have them provide the Proof that they need the memory, and be ready to add more.

They allocated 1.00 GB. Right there that tells me they just picked a 'nice round number' and are possibly clueless. If 0.80 GB is enough, then that's 200MB more for you free page list!

Make them do their homework and figure out how much memory is truly needed to make the Oracle part of the application work reasonably well.

If they claim 'it will go slower' then you respond 'really? Slower than "the system is hanging with 50 users connected."?'

Also, How much is your time worth, and the end user satifaction worth, compared to the price of increasing memory?!

Good luck!

Hope this helps some,
Hein van den Heuvel (at gmail dot com)
HvdH Performance Consulting

[Jur van der Burg]

You're not using XFC but the old VIOC cache (VCC_FLAGS is not set to 2). Why? XFC is way better. What VMS version are we talking about?

And why do you want a big freelist? You obviously need some free memory, but in essence free memory is wasted memory.

Install T4 on the system and let it run during prime time, that will give you someting to look at. Also, tinkering with memory parameters like freegoal and such makes no sense if you don't know the workload. Best is to leave them alone and let autogen do the job. Nowadays there's hardly anything you can gain with those parameters, except in special cases, but then you really have to know what you do, and you have to know the applications behavior.

Jur.

[Willem Grooters]

The problem is: If 50 users access the system, it hangs.
It wouldn't be caused by exhaustion of free space alone (that's what Virtual Memory System is all about: Using (far) more memory that is physical present in your system).

I concur with Hein: Login (as admin), have you users do their job, and start monitoring the system when 40, or 45, have logged in, add more users until the system hangs. Check available memory - and, most important, pagefile use.

If your free memory pool gets exhausted, the system will try to free pages from the mdoified page list by relocating the contents into the pagefile. If that get's exhausted, you may encounter a situation know as thrashing where the system will seem to hang - for all users.

If you could supply a SHOW MEMORY output just before this happens (say: when 48 users have logged in), that would show a better indication.

Since others have already indicated that your pagefile is rather small, it may be worthwhile to extend both page- and swapspace on disk (adding a new pagefile is easy and can be done without rebooting the system). But this can only be p[roven by the asked output.

Another possibility is that the software you use may cause a problem. What if the software that your users use, is accessing the Oracle dataase and exhasuting the Oracle dataspace?

[The Brit]

Remind me, you are running OpenVMS 7.2, but did you say whether you are on VAX or Alpha.

Other useful information would be the values of the PQL parameters.

vis.

sysgen show pql

I noticed that WSMAX is set to almost 1M pages, so it is quite important to know what the values of WSEXTENT and WSQUOTA are.

It is possible that processes are hogging the available memory as they login, and are not giving it back.

There is no automatic reclaimation taking place, (PFRATL = 0). Try turning this on.
(although it may not help since you dont seem to be doing any significant paging)

If you do turn it on (set pfratl > 0), suggested values are: (all dynamic)

pfratl = 1
pfrath = 10
wsinc = 2048
wsdec = 512

The fact that you are not paging may be a reflection of the fact that the processes which are "IN", are functioning with large Working Sets.

Just a couple of thoughts.

Dave

[Paul cotton]

Good morning,

I would just like to thank everyone for the feedback they've provided me on this issue to-date. I apologise for taking so long in responding to everyone but since posting my last update yesterday have been partly offsite and am playing catch-up with various things today

I may not get an opportunity to respond to each of your posts individually (but I'll try to respond to those which i feel are relevant/have requested feedback). In tne meantime I've tried to awward points as fairly as possible.


In terms of the actual issue i reported I will leave it open until next monday before deciding whether the thread can now be closed
Basically next monday is the earliest that malaysia can arrange to have live users reconnect to that system - currently using a backup system instead).


Also since reading your posts I've discovered that the oracle alert file on the DS25 system is indicating that the memory reserved for Oracle is set to higher value than required , which does indicate that Hein was correct - (so thanks for that feedback especially Hein), and that oracle are essentially tinkering with the sizing rather than actually calculate a correct value.

What I have also noticed is that the users working set quotas are set quite high so have enabled awsa and hopefully they'll get reclaimed if necessary. I'm also increasing the size of the pagefile
so that hopefully free page list will now be able to expand by transferring contents of modified page list onto the pagefile if required (I had increased mpw_hilimit a few days ago for this very reason but not had a chance to test whether this provided free page list with additional memory yet.

I'll try and come back to those people that have requested feednback from me but if not then I'll post an update next monday.. thanks for your time.

regards
Paul

[Paul cotton]

Good morning

I may have been a little bit over enthusiastic in awarding points on this posting. This is my the first posting on this forum, and having just read one of the alert notification emails from HP i now understand the way points should be awarded!

This issue is currently still open as I'm waiting on the guys in Malaysia to complete some associated tasks before deciding
whether this call can now be closed.

If the system continues to hang then I'll capture the show mem/full output + anything which looks relevant and post it on my next update.

[Guenther Froehlin]

Paul,

why is your focus on increasing FREELIM? It would make things worse by keeping an artificially long free page list. Let processes get to free pages.

When the system "hangs" what "hangs"? Certain applications?

You need to look at "SHOW SYSTEM" for any wait state. Like RWCSC, RWMPB... My wild guess is that the system ran out of page file space and processes got stuck in RWMPB (resource wait modify page writer busy).

A "SHOW MEMORY/FILES" from the loaded system would give an idea about the page file sizes.

/Guenther

[Hoff]

Free memory is (by definition) wasted memory.

The old Free Memory "stuff" changed at V7.2, with the advent of XFC. As has been mentioned here in the thread.

Some (now-old) performance-monitoring tools (erroneously) report configuration errors starting at V7.2 due to the changes in how system physical memory was utilized.

The points system serves as a distraction, and one of (many) ill-conceived and disruptive (and buggy) features of the user interface used here in ITRC.