[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

1251.0. "Notification problems" by COL01::WELZEL (Uwe Welzel, NaC, Cologne) Tue Jul 16 1991 08:09

Hello,

I'am running DECmcc here in Cologne. The configuration (Map)
has about 400 Domains (DELNI, DEMPR etc...) and 800 entities (Node4, 
Bridges, Station etc...).
DECmcc Ver 1.1 (BMS) including the following AM's
TCPIP V1.0.0 (SSB), Vitalink Ver X0.1.0, TSAM Ver T1.0.1, ICMP V 1.1 (french 
written AM) is installed on a LAVC running VMS V5.3-1:
DECmcc Alarms and TCPIP Dictionary fix patches are applied.
A MVAX 3800 with 32 MB and VS 3100 (Satellite) for MAP display.

Now, if we enable Notification on a Map the process takes a long time and
a lot of cpu power. (Will this change ?)

Then there comes a lot of error Messages:

-on the terminal where the DECmcc Interface was started
 SYSTEM-F-ACCVIO Access violation, reason mask=00, virtual address=000000C,
 PC=000D8180 PSL=03C00004
 (every time the same access violation)

-on the DECmcc window
 MAP Window Message: Notifaction being stopped for Domain Namesp_ns:.Domainname
 %MCC-E-UNSUPP_OP unsupported operation
 (for lot of Domains, but not all)
and sometimes the DECmcc window disappears

Has anybody an explanation?

Thanks Uwe

T.RTitleUserPersonal
Name
DateLines
1251.1RE .0 CompletionCOL01::WELZELUwe Welzel, NaC, CologneTue Jul 16 1991 14:2214
    re.0
    
    The access violation error only occurs if another process runs DECmcc
    i.e. Alarms Batch Job or User running interactive DECmcc.
    If I start DECmcc on the System alone and enble Notifacation at this
    time, no MAP error messages or access violation appears. After the
    notfication is enabled, I submit the Alarm Job in batch. If the alarm
    detect an error, the Notfication will be stopped for the Domain where
    the alarm was detect, with the same MAP error message in .0.
    But the Access violation error on the process screen, where DECmcc was
    started, did not appear.
    
    Uwe  
    
1251.2notification problems and where we are goingTOOK::CALLANDERJill Callander DTN 226-5316Wed Jul 17 1991 12:1033
My guess is that your 400 domains are probably all dynamic domains, and that
a significant portion of those domains contains a majority of the others.
I am curious to know if the entire 400 domains are setup in a single
hierarchy?!

When you start up the map, notification is started for each dynamic domain
within the domain that was started (all the way to the bottom of the
dynamic domain hierarchy). This, along with alternate identifier translation
are what is taking so long. For V1.2 some changes will be made to move some
of that processing down into the notification FM. Doing this will allow us
to put the request in an seperate thread, allowing the UI to continue while
the setup is on-going (user visible performance enhancement, but events won't
start coming through until setup is complete).

As to the accvio and errors, if you have the alarms error that would help
in figuring out what is going on. One potential problem could be memory.
I need to know how your domains are structure so as to estimate the number of
threads (and the amount of memory) that is being used to setup the notify
commands. Also if you have the time, I would like an FCL log of the
equivalent function.

$ deassign mcc_init
$ define mcc_fcl_pm_log 28
$ define mcc_notification_fm_log 8
$ manage/enter
MCC> notify domain xxx

You will need to do a notify command for each domain in the hierarchy (dynamic)
under xxx. Best to run it using a set host/log to capture it all.

thanks
jill
1251.3mail the log, don't post it, or send a pointerTOOK::CALLANDERJill Callander DTN 226-5316Wed Jul 17 1991 12:100
1251.4locking problemTOOK::HAOWed Jul 17 1991 14:3613
    About the IMPM problems when you have the alarms batch running before
    starting IMPM, there was another note somewhere in here that had a 
    similar problem.  I believe the problem is due to a bug in the IMPM
    where we are not locking global data structures that are shared.  What
    is most likely happening is that the map structures are still being
    initialized, and the notification code is attempting to read/write
    to those structures because notification is being received.  
    
    We are in the process of putting mutexes around these global
    structures, in time for v1.2.
    
    Christine
    
1251.5Similar problemCX3PT2::SHOTO::W_MCGAWMon Feb 24 1992 18:2220
I have a customer who was experiencing problems similar to those reported
in .0 and .1.  He had in excess of 100 domains that I believe are all dynamic.
Some of the domains were created just to represent pc's on the wan that
could not be registered.  They wanted representation of the pc's so information 
about the lcoation etc. could be stored.

They have removed the domains that were pc's and got the total down to 88.  Now
notification doesn't get stopped as it originally did.  He has even added some
of the domains back and is now at a total of 98.

Is there some magic number of domains at which point the notification fm can't
handle the load?  Is it based on the number of alarms also or is it just due
to the number of domains?

My customer is an internal DEC person working as a resident on a customer site
and could probably provide more information as needed.

Thanks'

Walt
1251.6use ref entities instead of domainsTOOK::CALLANDERMCC = My Constant CompanionMon Mar 23 1992 20:497
    it is based on the amount of memory you have. Each domain requires
    the notification FM to request gevents for that domain from the alarms
    FM. This memory usage adds up. Please use reference entities instead
    of domains for displaying non-manageable devices.
    
    jill
    
1251.7Actually we are... nowCX3PT3::SHOTO::W_MCGAWMon Apr 06 1992 22:4110
    Reply to .6
    
    Is there some way to increase the amount memory that DECmcc has
    available in the event that all of the domains are true domains and not
    being used for keeping track of the pc's?  Can I check to see how much
    memory is being used for notification?  Could you also expand on
    'reference entities'?
    
    Thanks
    Walt
1251.8notification scaling being worked.TOOK::CALLANDERMCC = My Constant CompanionMon Apr 27 1992 23:0414
    please be patient with us, we are in the middle of reworking the
    notif/alarming interaction so that it scales better for the larger
    (read that as alot of domain) environments. We should have this
    resolved shortly so that this becomes a non-issue for you.
    
    As to reference entities, this is a new class of entity that was
    defined as a "place-holder" for non-manageable (either because they
    really are non-manageable or just because you don't want to manage
    them). You can maintain reference information on them but nothing more.
    They are useful for drawing pictures of your environment, or
    maintaining inventory information on miscellaneous hardware.
    
    jill