My point is that it is good to have an empirical performance comparison (which would mean everything is apple-to-apple including setup, use case etc.) between Netconf and FORCES given that both the protocols have been in existence for a while.
I agree that there may be some flaws in this particular comparison between SNMP and Netconf. I couldn't find anything else which is easily publicly available; if you have any better publicly available reference, please suggest. Thanks, Ramki -----Original Message----- From: Jamal Hadi Salim [mailto:[email protected]] Sent: Tuesday, April 22, 2014 7:15 AM To: Juergen Schoenwaelder; ramki Krishnan; Joel M. Halpern; Andy Bierman; Jamal Hadi Salim; Russ White; [email protected]; Jan Medved (jmedved); Dean Bogdanovic; Edward Crabbe Subject: Re: [i2rs] consensus on I2RS protocol and model I see a different flaw just from the first few pages and looking at the conclusion. Sorry - didnt have time to look at the whole thing. The paper picks a model entity of about 90% strings (theres one "int" iirc). I am not sure what that was supposed to emulate. From that perspective the question answered to me seems to be "how bad does SNMP do when you have mostly strings?". It seemed to do fairly well at the end. I liked the fact they tried to batch things with SNMP. If i was emulating what I2RS is doing by building on say the RIB model, it will not be 90% ascii on the wire for SNMP. Agree with Juergen on apple-apple comparison needed. cheers, jamal On Tue, Apr 22, 2014 at 3:19 AM, Juergen Schoenwaelder <[email protected]> wrote: > On Mon, Apr 21, 2014 at 10:26:58PM -0700, ramki Krishnan wrote: > >> It would be good to have an empirical performance analysis of Netconf >> vs FORCES; this would substantiate performance advantages, if any, of >> either protocol. The following reference provides a reasonable >> empirical performance analysis of Netconf vs SNMP - >> http://morse.colorado.edu/~tlen5710/11s/11NETCONFvsSNMP.pdf. > > I believe this comparison is flawed. ncclient uses a pure Python > implementation of SSH while the authors used a Python wrapper around > NET-SNMP (a C implementation) to measure SNMP performance. And the > authors state that they used SNMPv2c (no security) and SSH as the > NETCONF transport. (It is also unclear whether all NETCONF > transactions were executed over a single session or multiple > sessions.) > > The design of such experiments is not as easy as it may look at first > glance and it is in particular important to use comparable security > algorithms (ideally the same crypto library). > > /js > > -- > Juergen Schoenwaelder Jacobs University Bremen gGmbH > Phone: +49 421 200 3587 Campus Ring 1, 28759 Bremen, Germany > Fax: +49 421 200 3103 <http://www.jacobs-university.de/> _______________________________________________ i2rs mailing list [email protected] https://www.ietf.org/mailman/listinfo/i2rs
