Operating System - OpenVMS
1830107 Members
29101 Online
109998 Solutions
New Discussion

Performance DECREASE after CPU/memory upgrade

 
Richard W Hunt
Valued Contributor

Re: Performance DECREASE after CPU/memory upgrade

A test for which the results have not been fully explored is the output of MON STATES.

If MON MODES or its equivalent shows you that you are running the CPU a lot, up to 100%, then your gerbil is running as hard as it can. However, is all of that USER mode or is some of it in the more elevated modes?

I am thinking that if you didn't run AUTOGEN your page/swap file sizes didn't change even though you doubled memory. This might radically change how the system responds. So the question isn't whether your CPU is saturated but whether some of the saturation is for states other than COM/CUR. Some of those other states are treated as non-idle, too. Only HIB and LEF and SUSP (and their "O" variants) plus the MWAIT and its variants are truly idle. Further, if MON MODE shows 100% usage, that doesn't always mean it is your task getting the usage. I think you need MON PROC for that.

Another thing to consider is this: If you boosted memory, did you boost your process quotas to prevent in-memory thrashing? I.e. lots of page faults satisfied form the MPL or FPL? Check MON PAG for an increased Free List Fault Rate or Modified List Fault Rate.

You would think that with more memory, you would always gain - but so many of the system and per-user parameters are so interdependent that it is sometimes hard to be sure.
Sr. Systems Janitor
Willem Grooters
Honored Contributor

Re: Performance DECREASE after CPU/memory upgrade

Well, I DID run autogen (both when up- and downgrading), and in both cases the pagefile was changed (increased/decreased in size) - I don't have a sawpfile. I did not change memory parameters explicitely, left that to autogen, which in turn changed the settings for global sections and -pages, but not for process-bound quota (as far as I know).
When I observe the system (so my process is shown in CUR state), this CPU-bound process is obviously in COM state. The rest of the ~ 50 processes are HIB or LEF - so really idle. Would I need to adapt the process-bound process quota as well to increase that speed? I would need to know the memory requirements of each of them - and how to get these when these are (mostly) idle....But I'll see what I can get out of the system. (Let moitor run for some time, say 24 hrs (or longer) ?)
(although specified otherwise, I would really like an answer since i 'need' that extra mamory...)
Willem Grooters
OpenVMS Developer & System Manager
Richard W Hunt
Valued Contributor

Re: Performance DECREASE after CPU/memory upgrade

Looking back through the notes, I saw something else that I'm not sure about.

Your dump of the output that shows your memory modules shows 3 modules. You cannot interleave symmetrically with three modules. You need two or four. If you have disrupted memory interleaving somehow, that might account for a speed difference.

Otherwise, I have to say I'm stumped.
Sr. Systems Janitor
Martin P.J. Zinser
Honored Contributor

Re: Performance DECREASE after CPU/memory upgrade

Hi Richard,

each of the "banks" is two slots for a total of 6 in the PWS, so interleaving should not be the problem.

Greetings, Martin