[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference azur::mcc

Title:DECmcc user notes file. Does not replace IPMT.
Notice:Use IPMT for problems. Newsletter location in note 6187
Moderator:TAEC::BEROUD
Created:Mon Aug 21 1989
Last Modified:Wed Jun 04 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:6497
Total number of notes:27359

2411.0. "CPU/Memory requirements" by CLARID::HODSMAN (The Giraffe) Tue Feb 25 1992 05:06

    I am running DECmcc t1.2.4 on a 16MB 3800 a 4.2 VUP machine.
    I understand that 16MB is a little light.
    
    When I start 40 alarm rules (all OCCURS) 
    a monit proc/topcpu shows that on average 14% of the CPU
    is taken just by process that runs MCC_main.exe.
    
    This seems a large amount of CPU power just to keep the alarm rules
    enabled.  Is it normal ?  If not is it AM related.
    
    I am aslo suprised that I cannot create and enable 70 alarms rules
    on this machine - process threads abort, and I have get messages
    peratining to insufficient dynamic memory.
    
    Would it be possible to have an authoratitive figure on how much
    memory and CPU is used for each alarm rule?
    
    If I wanted to use 1000's of alarm rules would I be better off
    switching the management agent to use data collector.
    
    Regards,
    Jeremy
T.RTitleUserPersonal
Name
DateLines
2411.1An educated (?) guessNANOVX::ROBERTSKeith Roberts - DECmcc Toolkit TeamTue Feb 25 1992 07:2314
RE: .0

>    Would it be possible to have an authoratitive figure on how much
>    memory and CPU is used for each alarm rule?

  I haven't done an extensive research, but an Alarms Rule consumes about
  20k bytes of memory for the internal data structures.

  Then there is stack space for the thread to use:  about another 20k bytes

  Then there is Thread overhead; memory consumed by the threads package
  to maintain the thread:  another guess of about 20k bytes.

  /keith
2411.2120-200 pages/rule is a good guessHAZARD::BAKERPaul Baker, UK Product and Technology Group - 844 3311Tue Feb 25 1992 16:4117
    
    I would agree with Keith's estimate. I have seem guesstimates of
    between 120 and 200 pages of memory per rule, depending on whether it's
    sitting there waiting for an event or actively going out and polling.
    
    Also, I have seem my 16Mb V5.4-3 system die with upto 800 page
    faults/sec running DECmcc. The "SPD" says 24Mb as a minimum - I thing
    it's correct. If you want decent performance, go for at least 32Mb.
    
    Now - what we really need before V1.2 ships are the real numbers that
    we can quota to a customer when sizing systems. All this guessing is
    not going to get us any friends in the customer community if we get it
    wrong.
    
    Paul.
    

2411.3Some performance enhancements for Alarms will be in v1.2NANOVX::ROBERTSKeith Roberts - DECmcc Toolkit TeamTue Feb 25 1992 16:597
  V1.2 of Alarms will have some memory consumption fixes.  Memory which is
  needed during evaluation - but not during hibernation was not freed.  This
  will be fixed and should save anywhere from 1k and 4k per rule.  Every
  little bit helps.

  /keith
2411.4and cpu usage?CLARID::HODSMANThe GiraffeWed Feb 26 1992 02:339
    Thnaks for the news about memory - any news about cpu usage
    which is my real worry.
    
    15% of a 4.2 vup machine to keep 40 alarm rules going 
    suggests 15000 instructions a second just to keep an alarm rule
    enabled.     Is this normal - do I need alpha already ?
    
    regards,
    jeremy
2411.5There's a lot of stuff going on inside the boxNANOVX::ROBERTSKeith Roberts - DECmcc Toolkit TeamWed Feb 26 1992 08:4824
RE: .4

>> ... 15% of a 4.2 vup machine to keep 40 alarm rules going suggests
>> 15000 instructions a second ...

  There are a lot of things going on inside.  The Alarms FM calling
  the Access Module (DECnet, Bridge, SNMP, ...) must go through the
  Information Manager; Scheduling algorithms are performed and dispatching
  takes place.  Thread context switching and overhead eats up some CPU.
  If Notification and the Iconic Map are enabled then more threads
  are executing -- and events produced by Alarms are processed through
  the Event Manager, which needs some CPU too.

  The DECthreads (CMA) package we are using for v1.2 seems to consume
  a bit more CPU and memory too (more than the MCC threads package used
  for v1.1)

>> do I need alpha already ?

  In many respects the Alpha chip would fix all of our problems.  But
  certain DECthread internal implementations just consume CPU (semaphores
  I believe poll for events); A faster machine may not be the only answer.

  /keith
2411.6HW >>>MAYDAY::ANDRADEThe sentinel (.)(.)Thu Apr 09 1992 09:4514
    I understand how a product can get a bit out of hand when its first
    developed. And was given to undertstand that things would get better
    when version 1.2 came out. "The performance would noticibly improve"
    
    But what do I hear now, v1.2  actually consumes more memory and cpu
    then v1.1 did  ????????   How come, did the added functionality eat
    away all the performance gains ?
    
    Are there plans to improve DECmcc performance for v2 ? (for real)
    
    *** DECmcc *** the only product to regulary quit on me due to lack 
    resources.... 
    
    Gil
2411.7catch 22, but we are working on it.TOOK::CALLANDERMCC = My Constant CompanionMon May 11 1992 17:2412
    
    don't give up yet. We are having some problem with some of the
    underlying tools we are dependent upon. But they are being VERY 
    responsive with fixes (memory and functionality). We are also working
    full time on fixing performance and memory problems as we speak. If you
    are interested in more details on this please call me and I can discuss
    the changes that we are making. But for now surfise to say that we are
    well aware of the problems. The biggest thing for us was that we did
    enhance the memory usage and performance, but then went and added in a
    ton more functionality which just went right back and consumed it all
    back up again....