On Tue, Apr 8, 2014 at 4:29 PM, Alison Chan <chan7...@kettering.edu> wrote:
> I was able to take the router back up to the lab (I had been testing > it with laptops in my office). With two gigabit ethernet hosts on the > data plane, throughput (tcp iperf) is 450~460 Mbit/s. I will test > things a bit more tomorrow (e.g. adding 802.11 clients) and if all > goes well then I will move all five of our TP-Links to OVS. > There are two variants of this device - one with an AR9132 (boring white case), and one with a QCA9558 (blue spaceship-looking thing). The AR9132 version uses a separate switch ASIC (AR8316 - I have one of these), so your max throughput through that device in any software forwarding (regardless of CPU speed) will be a gigabit. The 400Mhz ARM in that box is going to limit you even further, so 450Mbits seems a bit suspect. If you have the rev. 2 (blue) hardware, that is more of an unknown (I don't have one of these to look at or tinker with) - the 9558 has a 720Mhz MIPS core, so that's better, but I found images of a different AP that uses this chip on the internet and it appears to use an AR8327 for the switching, so likely there's no integrated ethernet switch on the SoC. This means your max throughput is still probably a gigabit, and you might be able to actually get 450Mbits through that CPU. The AR8327 actually has a 96-entry ACL table and 64-entry VLAN translation table (as well as a bunch of more specialized tables that would be more difficult to use well) that could be used for hardware acceleration, but I doubt anyone has done the work to program them. -- Nick
_______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
