Thanks Nick for replying In the first question I was referring to the flowmod messagesreceived from the controller (not actual flows) that a hardware switch can process. Does anyone have a rough approximate of this value for a hardware switch? (Just curious) Is there any tool for measuring theperformance of a hardware switch ? When can we expect support for version 1.1 by vendors? I am wondering if there are some chips which have hardware support for1.1 version or new chips need to be manufactured for this version.If latter, I think it would take some time right? I suppose there would have been discussions with chipmanufactures prior release of ver 1.1 on the feasibility of certain actions.But first need to see if some vendor has full support for version 1.0. Karthik
--- On Sat, 4/30/11, Nicholas Bastin <[email protected]> wrote: From: Nicholas Bastin <[email protected]> Subject: Re: [openflow-discuss] Questions on hardware switches To: "karthik" <[email protected]> Cc: [email protected] Date: Saturday, April 30, 2011, 3:12 AM On Sat, Apr 30, 2011 at 02:15, karthik <[email protected]> wrote: I have some questions on openflow mainly related to vendor hardware switches 1. The open flow controllers seem to be getting more robust and powerful. I was wondering how many flows/sec can a hardware switch (Hp Procurve,Indigo etc) process per second. Do you mean flowmods, or actual flows through the switch fabric? Generally the hardware switches can forward at line rate through any number of flow mods (the limiting factor usually being the size of the flow table, not the processing speed). Obviously you should refer to the vendor specs for more precise information. 2. Can the controller overwhelm a hardware switch by sending packet out ,flow mod and flow stat commands? Very likely. If it can, does openflow protocol have some mechanism where the switch and controller can exchange the limits of its processing capabilities.Example if a switch can handle only say 50 flow mod msgs/sec it can communicate to the controller at the initial handshake about this and controller restricting the flows to this limit. This is outside the scope of the spec - every device will behave differently and while controllers could encode this information, it's certainly not codified in the spec (different requests will have unknown requirements on CPU time, it would be hard to communicate this in a handshake). Last question, how many vendor switches support openflow now (ver 1.0 and 1.1) I don't believe any available hardware switches support 1.1 (at least not publicly) - this spec is very new and they've barely had time to read it, let alone implement it. The only commercially available switches that I know of that support openflow out of the box are from Pronto and Netgear (there may very well be others I don't know about - don't count this list as exclusive by any means), although there are various other vendors who have experimental firmwares that you can likely acquire if you communicate with their sales staff. --Nick
_______________________________________________ openflow-discuss mailing list [email protected] https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
