[ajb]

I do, thanks.Hmm, let's see if I can give you the information without breaking a confidentiality agreement. Ah, here is a discussion on the openwrt mailing list: https://lists.openwrt.org/pipermail/openwrt-devel/2013-Augus... As you can see, they reported 90mb/s (they also quote 710, but I think that is the switch doing that). And that's on a mips74k.

I know it's made for routers, but most likely not for doing a lot in software. As you quote, it does hardware NAT. There's a reason they put that in.

There isn't a fundamental reason why the mips couldn't do the job, but running stock linux it's going to be thrashing its L1 cache, and it doesn't have an L2. Also, next time you play with one of those,check the hardware counters for how many instructions/sec it actually manages to process running flat out :-)

[listic]

90Mb/s at least? Makes sense that they won't go for the gigabit Ethernet. Any idea how that compares to ARM with the same frequency? I'd say about the same. That's not bad, right?

[ajb]

I would guess about the same; I didn't have two similar ones to compare. The mips might be a bit slower though. I had the sense that the limitation was really L1 thrashing of code space, so the ARM being a bit denser might help. It would also help if the SOC had a smaller latency to memory as well - don't know what this one has.

Bear in mind that it's probably a packet rate rather than a bit rate limitation. That really bit us in our application because we were trying to run many lines of voip, which means lots of small packets. - YMMV.