[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

4260.0. "strange behaviour of SNMP AM " by ANNECY::HAGENMULLER (EIC Annecy - dtn 7887.41.99) Wed Dec 16 1992 05:32

This note pinpoints a problem we have with an SNMP proxy agent
(still the Stratacom's one for those who regularly read this conference ...).

When trying to get statistic data for a frame realy port, DECmcc seems to take
a lot of time (about 30 s for 17 attributes (the counters partition)). Part of
this delay rests upon the proxy agent itself which is performing some odd
calculations to genetate the values, but a non-small part falls also to 
DECmcc.
At least the enclosed mail outlines a non efficient behaviour of the SNMP AM.
Is there a work around ?

Christophe.
T.RTitleUserPersonal
Name
DateLines
4260.1mail from StratacomANNECY::HAGENMULLEREIC Annecy - dtn 7887.41.99Wed Dec 16 1992 05:33128
                  I N T E R O F F I C E   M E M O R A N D U M

                                      Sent:         GMT
                                      From:       
                                                  [email protected]@umc@WRLMTS@WRL
                                      Dept:        
                                      Tel no:      
                                      Message no: 000798
                                      Reply to:   

TO:  hagenmuller@a1_annecy.valmts.aeo.MTS@umc@WRLMTS@WRL


Subject: Multiple Request Problem


Christophe,

    This is a message from our engineer on a problem found with DECmcc on
the multiple OID request for the counters.  Please let me know what you
think.

							Regards,
	
							Clyde

=======================================================================

>From ts Mon Dec 14 12:58:03 1992
Date: Mon, 14 Dec 92 12:58:02 -0800
>From: ts (Timon Sloane)
To: cki
Subject: DECmcc bugs

Clyde,

I have diagnosed the "multiple attribute request" problem
and it looks like a DECmcc problem.  The following is
what the trace looks like when trying to dislay the "Counters"
from DECmcc of one of the packet lines.

Indices 12 through 33 are the Gauges in the MIB.

KEY:
	-> GET 12-26 
		means DECmcc requested indicies 12 through 26

	<- GOT 12,26, noSuchName #15 
		means a GET response was returned to DECmcc with a 
		noSuchName error on index #15.  SNMP protocol
		requires in these cases that no valid data be returned
		for any of the other variables requested in the GET.


_+_+_+_+_+_+ TRACE _+_+_+_+_+_+_

1  -> GET sysUptime .. sysObjectId
2  <- GOT noSuchName
3  -> GET 12-26
4  <- GET 12-26, noSuchName #15
5  -> GET 27-33
6  <- GOT 27-33
7  -> GET 12-14, 16-27				ALREADY HAS 27!
8  <- GOT 12-14, 16-27, noSuchName #16
9  -> GET 28-33					ALREADY HAS 28-33 !!
10 <- GOT 28-33
11 -> get 12-14, 17-28				ALREADY HAS 27 and 28
12 <- GOT 12-14, 17-28, noSuchName #17
13 -> GET 29-33					ALREADY HAS 29-33
14 <- GET 29-33
15 -> GET 12-14, 18-29				ALREADY HAS 27-29
16 <- GOT 12-14, 18-29, noSuchName #18
17 -> GET 30-33					ALREADY HAS 30-33
18 <- GOT 30-33
19 -> GET 12-14, 19-30				ALREADY HAS 27-30
20 <- GOT 12-14, 19-30, noSuchName #20
21 -> GET 31-33					ALREADY HAS 31-33
22 <- GOT 31-33
23 -> GET 12-14, 19, 21-31			ALREADY HAS 27-31
 ** 5 second timeout **
24 -> GET 12-14, 19, 21-33			ALREADY HAS 27-33
25 <- GOT 12-14, 19, 21-33
26 -> GET 32-33					ALREADY HAS 32-33
 ** 5 second timeout ** (proxy is processing #24)
27 -> GET 32-33					ALREADY HAS 32-33
 ** 5 second timeout ** (proxy is processing #24)
28 -> GET 32-33					ALREADY HAS 32-33
 ** 5 second timeout ** (proxy is processing #24)
29 -> GET 32-33					ALREADY HAS 32-33
 ** 5 second timeout ** (proxy is processing #24)
30 <- GOT 12-14, 19, 21-31
31 <- GOT 32-33
32 <- GOT 32-33
33 <- GOT 32-33

_+_+_+_+_+_+ TRACE _+_+_+_+_+_+_
Note that DECmcc does a few things that are very inefficient.

First, it requests the same variable multiple times.  This looks
like a serious and obvious DECmcc bug.

Second, once it has discovered that the table is sparse, it continues
to try to build large requests.  Instead, it should back down
to small requests (with 1 or 2 variables) because in sparse
table situations large requests are much less efficient.  Take for
example the request response pair:

	3  -> GET 12-26
	4  <- GET 12-26, noSuchName #15

This transaction requires the agent to calculate the values for 
12-14 before it discovers that 15 is not available.  The response 
(msg #4) does not contain the values for 12-14 (that is the way 
SNMP works).  So all the time calculating 12-14 is wasted.  Smaller
requests avoid this problem.

Third, I had reset the UDP timeout to be 60 seconds.  I was still 
seeing 5 second timeouts on the wire.  It looks like there is 
an ordering dependency between adding an IP device or opening an
IP device and resetting the UDP timeout.  DECmcc also does not
seem to save the new timeout value over restarts.

timon


=======================================================================

4260.2DEPENDS ON managed by PM and not AM ?ANNECY::HAGENMULLEREIC Annecy - dtn 7887.41.99Wed Dec 16 1992 09:1226
One way to restrict the number of requested attributes would have been to use 
the "DEPENDS ON" qualifier in the MSL defining the counters attributes.

I modified manually the MSL produced by the MIB compilation and applied some
rules like "if the chracteristic attribute XXXX has an enumerated value of ...
then counters attributes ..... are not defined" by using the "DEPENDS ON"
qualifier when defining the counters attributes.

But every time I ask for the counter partition, DECmcc seems to send a SNMP 
request containing all the counter attributes and filters the attributes only
at the PM level (one get an error message which looks like "extra info was
returned for snmp .... : code ...."). Am I right ? Or is there a way to
reduce the number of attributes actually requested on the network (by excluding
the attributes whose value is known to be undefined) either with a "DEPENDS ON"
qualifier or another mechanism (seems that the notion of group doesn't help
since it can't be accessed from the iconic map and doesn't prevent DECmcc
to query all the attributes pertaining to the same partition as the attributes
in the group ?)


Ch.




4260.3It is a known weakness in the SNMP AM.2582::YAHEY::BOSEWed Dec 16 1992 09:399
	Christophe,

		The behaviour that your customer saw is already known
	to us. It is on the list of things to do in SNMP AM for the next
	version. The fix might go into V1.4 (the version after V1.3) based 
	on availability of resources.

	Rahul.