| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1245.1 | varies; configuration-dependent. | MCDOUG::MCPHERSON | i'm only 5 foot one... | Mon Jul 15 1991 09:36 | 35 |
| re .0
> 1) Is there a limit of the number of objects a single DECmcc can
> handle, and when (if) we have director to director communications will
> the maximum number of objects handled be (n * max. objects) where n is
> the number of directors and max. objects is a constant ?
I believe this question has been asked two or three times before in
this conference (although I don't remember the answer, sorry). Try
doing some DIR/TITLE=_____
I assume that you're asking for a *practcial* limit and not a
theoretical limit. The practical limit is configuration-dependent, so
"your mileage will vary". Maybe someone in MCC development could post
the *theoretical* limit, if there is one.
> 2) If a live DECmcc is polling the network for alarm procedures, how
> many polls does it have to initiate at one time to have an effect on
> the ethernet. Is there a quantifiable value ?
As opposed to a dead DECmcc? ;^) The amount of ethernet traffic is
dependent upon the class of entity being polled and what attribute
partitions are being examined. If there are few attributes (or data
values) in the partition(s) being polled for, then there would be
little traffic per poll. If the entity you're polling has a *lot*
of counters and you're only looking for *one*, you could generate more
traffic than you expected. Note that traffic is coming *back* from the
managed entity, not from the director.
Once again, this is *extremely* configuration-dependent, so I wouldn't
hold my breath fpor any sizing estimates...
/doug
|
| 1245.2 | your answers were fine | TOOK::CALLANDER | Jill Callander DTN 226-5316 | Mon Jul 15 1991 16:16 | 3 |
| as to the theoretical answer to one, the limit is an infinity number
of classes. but, in reality memory and cpu limits will always have
an impact on the actual limits.
|
| 1245.3 | Thanks on 1)...., but 2) ! | WOTVAX::PURNELLR | ...all this now ! | Mon Jul 15 1991 17:08 | 23 |
|
doug,
Thanks on question 1).
Is there any way of quantifying impact on ethernet with polling. As the
enterprise a DECmcc manages increases, impact on the ethernet (or
whatever physical transport is used to convey information between an
object and a director) will become a greater issue.
Are any figures, guidelines going to be offered ? This question I have
received several times......customers usually trying to imagine the
impact of having a singular as opposed to multiple management platform.
I shall pose another question, is there a DECmcc cookbook (ie useful
hints for installation and implementation of a station) either existing
or in the making ?
;*(
Rex
|
| 1245.4 | | MCDOUG::MCPHERSON | i'm only 5 foot one... | Mon Jul 15 1991 17:16 | 15 |
| 1. get a sniffer
2. Poll
3. count response packets
Repeat steps 1-3 for different attribute partitions and classes of entities.
Pick some averages normative values.
Run everything through your favorite AI-based multimedia document processing
application and...
Post your final results in here so all we have to do is point subsequent
requests for this data to your note. 0.5 * 8^) <-- (only 1/2 kidding)
/doug
|
| 1245.5 | your milage will vary | ENUF::GASSMAN | | Mon Jul 15 1991 19:45 | 23 |
| There are several issues to consider. Polling traffic will effect the
ethernet, but since the ethernet can handle 18,000 packets/second, it's
doubtful that you will burn up too much of the capacity. Beating on
the remote agents could be a problem though. Many devices have limited
resources for management, and hitting them every few seconds could
cause a problem. MCC's DECnet AM's technique of establishing a new
logical link every time it polls the same node is a case where perhaps
another way would have reduced the load on the remote system. Another
thing to look at is how the manager does availablity checking. Can the
router be asked for status of all 'end systems', or must each 'end
system' be polled. I think that in this case, DECmcc can poll the
router, but has problems flagging the correct system with the
reachability problem.
The ability of DECmcc to record attributes and counters can help reduce
polling frequency for some kinds of operations - such as finding out
how many level 1 routers you have - etc..
One limiting factor in a management system is it's ability to poll
quickly. Performance factors for DECmcc are not all in yet on this
attribute (would be a nice attribute for a manager of manager agent).
bill
|