Conference azur::mcc

4938.0. "IP poller problems" by ZUR01::SCHNEIDERR () Thu Apr 22 1993 11:33

Hello,

We run MCC 1.3 on a ULTRIX 4.2A system (5240, 96 MB mem, 3GB disk). We have a
realy complicated domain structure and we have some problems with the IP poller.

Our domain structure:
We have a CICSO brouters with 21 interfaces. each interface is registered as
a SNAM node. All this nodes are in a domain called CISCO_1, CISCO_2... for
each brouter. This CISCO domains are in different domains as member. Excactly
in 20 domains.
See the diagram....



				MGB
				 |
			-------------------
		        |		  |
                   building_1         building_2  .........
			|		  |
   -----------------------------
   |		|	       |
CISCO_1     floor_1	  floor_2 .............................
		|	       |
              CISCO_1        CISCO_1 

You see the same CISCO is on different domain levels. This diagramm is only a 
part of the hole structure.

Now the problems:

1. If you enable IP poller on level MGB, then all enities in all domains, in the
subdomains, too, are polled. I thought, only the entities in the related domain
should be polled.  -----> BUG?

2. On the level building_x some IP nodes (CISCO interfaces) are registered, too.
If you want to stop IP poller on level MGB, it does not stop all polling. The
only way is to kill the process. It seems, that only the nodes in the subdomains
stopped, but all double registered entities in the topdomain do not stop!!!!
---> BUG?

3. We poll about 100 nodes. We can see with the analyzer, that the ICMP packets
travel away well. The ICMP answers arrive well, we can see it on the analyzer.
They arrive in about 2-6 ms. The ULTRIX machine catches all ICMP answers. We can
see it in the counter SNMP xxx ICMP icmpInEchoReps. But the MCC don't catches
all echos. IP poller makes retries. We had to choose 3 retries to catch all
echos.
---> What parameter we can adjust to catch all echos from MCC, too?
---> BUG?

4. If one CISCO interface fails we get 20 notifications, because our CISCO 
domain, with the interface in, is member of 20 further domains. But all the
notifications tell us "from domain CISCO_1". But out entities are only member of
one domain, the CISCO_x domain.
---->BUG?



Thanks for helping

Roland    lot of questions, still :-)

	Here's some of the answers to your questions :

	1) When you enable polling in a domain, all entitities in the sub-
	   domains are also polled. This behaviour is intentional. It was 
	   thought that allowing this would mean less work for the user who
	   need not open each sub-domain domain and enable polling.

	2) This is probably a bug. Please file a QAR.

	3) Seems like the receive buffer for the socket might be dropping
	   some of the packets. The default buffer size is 4096 bytes and
	   needs to be increased. I will provide you with another executable
	   and you can tell me if there is any improvement in behaviour.

	4) I hope someone who is more familiar with notification will
	   answer this question.

	Rahul.