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

Re: why i should like pibs




[ Based on Dave Durhams comments, I will update my attempt to
  summarize the major technical contributions of COPS-PR over SNMP.  I
  will do so on the <rap@ops.ietf.org> only so that we can stop these
  cross posts. Below are just my hopefully final comments on Dave's
  numbers. ]

>>>>> Durham, David writes:

Dave> Actually, I did some pings and found my 10ms RTT was
Dave> unrealistic. 100ms is more like it. At 100ms RTT, things get
Dave> much worse for the SNMP equation for the configuration example I
Dave> described:

Dave> SNMP=2026 Seconds (over half an hour!)  COPS=2.59 Seconds

Dave> Now we're talking 1000x faster...

I probably do not understand your equations. There is the actual
transmission time Tx and the propagation delay Tp. We have

     n = number of filter entries
     m = size of a single filter entry
     s = "speed" of the link
     r = round trip time

If I understand you correctly, you assume:

     Tx_s = (n * m) / s		# SNMP n messages of size m
     Tx_c = (n * m/10) / s	# COPS filters are 10 times smaller

     Obviously: Tx_c = 1/10 Tx_s

You also seem to assume:

     Tp_s = (2 * n) * r		# SNMP sequential one set a time
     Tp_c = 0			# COPS does not need a round trip

In case this is your model (forgive me if I messed it up), I think it
is an over-simplification and the numbers you get out of it do not
impress me that much. (Did you ever calculate how big your TCP window
must be to do your example in one^H^H^Hzero RTT?)

Still I agree (and have ever done so in the past) with the qualitative
statment "COPS-PR is faster due to larger transactions and the TCP
transport in cases the network is running normally".

Dave> Remember also that COPS-PR presents a transactional model to the
Dave> user, while SNMP simply provides a get/set interface, so it's
Dave> not just an implementation issue; it's a presentation issue as
Dave> well.

I never got that into my brain. An SNMP SET is a single transaction
just as a COPS-PR DEC is a single transaction. The difference is just
the message sizes you can use.

/js

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