[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: IETF 55 EOS WG Minutes and Presentation




>>>>> Wijnen, Bert (Bert) writes:

Bert> I did not get a warm feeling at that point. In fact I was quite
Bert> disappointed with the response as to how many want to "help",
Bert> "review", "implement", "use" this technology.

I certainly like to use and support "a faster retrieval mechanism".
Retrieving SNMP tables of non-trivial size often reminds me of the
days where I found kermit over a 9.6k serial line a cool thing to
have.

I ship megabytes of documents around with virtually no delay
(according to human think times) on the Web these days, but getting a
few hundred rows out of the average SNMP enabled router nearby is so
damn slow that you can't really build serious applications other than
debuggers (called MIB browsers) or relative static polling engines.
Even answering simple questions such as "tell me the physical port and
the status of the interface with IP address X" by joining the
ipAddrTable, ifTable and entPhysicalTable on some of my real-world
boxes (in what people classify as an enterprise network) reaches
response times that are hardly acceptable in the 21st century.

I encourage Wes to move on and especially to implement his ideas as
part of NET-SNMP so that the SNMP community can play with it. If the
new protocol operations turn out to provide significant speedups for
things like the example query mentioned above, then I believe that
others will use these features in their applications and finally
customers will ask for applications that give a much more reasonable
response time.

Of course, one can argue that answering queries like the one mentioned
above with SNMP is not a good usage of this kind of protocol in the
first place. And in fact, on some boxes it is faster to just dump the
configuration and extract the relevant information from there. If the
consensus is that this is how we approach network management problems
in the future, then of course we do not need things like EOS and SNMP
is just fine as it is today.

/js

-- 
Juergen Schoenwaelder    <http://www.informatik.uni-osnabrueck.de/schoenw/>