Hi Panagiotis, Find my input inline:
On Tue, Jan 29, 2013 at 11:12 AM, Panagiotis Georgopoulos <[email protected]> wrote: > > This is what we are also seeing at the moment. For our experimentation we > could remove the L2 bits for now, however we would need to push certain > packets out of multiple ports which will essentially push us to software > flow processing again :-/ This is a tough situation. Very few of us are happy with how different vendors are putting different types of traffic into SW vs HW flows. I hold a belief that all flows should go into HW until you run out of HW space, then flows should go into SW. At least in that scenario you can limit the # of flows you enter into the box and not hit performance issues. I am even more concerned when it comes to multiple matches. But we will address that when we get there. >> Regarding the 1 Mbps limit, it's actually a packet-per-second limit that's >> configurable but defaults to 1000. Depending on your firmware version the >> upper limit may vary, but it can be as high as 10000 PPS; however, I am > not >> sure how well the CPU on your switch would hold up with that limit. You > may >> also be able to configure the openflow instance so that only hardware > flows >> are accepted, but I don't use that configuration; there are some flows > (ARP >> handling, for example) that work fine in software even with the PPS limit. > > This is really good to know; we'll give the PPS option a try and see if this > helps at all. Any CPU (Software Forwarding) is going to be bad and have random latency based on what other processes are running on the main CPU. >> Basically, in order to get reasonable performance you must keep the flows > in >> hardware, and in order to do that you must restrict them to just the > matches >> and actions described in the manual. . . >> >> Bill. > > I don't know if it is just me, (and I know that this is hard to get right) > but I would have been expecting more flows to be handled by hardware on the > 3500yl HP switch. I'd have to have a look around to see how much do other > vendors implement on hardware (any suggestion would be much appreciated). Pica8 handles > 4k flows (I tested up to 4700 or so) on their lower end (329x) switches and I believe up to 16k in their upper end switches (3920) (I can't find a link). The difference between Pica8 and some other vendors is that Pica8 has a OVS mode where you don't boot their normal PicaOS and just use OVS to run the switch. This allows for a lot more flows. In Hybrid (PicaOS) mode, the numbers are lower, around 1700 if I remember correctly. This is the right number to compare with fully hybrid switches such as the HPs. _______________________________________________ openflow-discuss mailing list [email protected] https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
