Date: Sat, 1 Mar 2014 09:05:15 -0800 From: Adrian Chadd <adrian.chadd@gmail.com> To: Ilya Bakulin <ilya@bakulin.de> Cc: "freebsd-hackers@freebsd.org" <freebsd-hackers@freebsd.org>, Alexander Motin <mav@freebsd.org>, "freebsd-arm@freebsd.org" <freebsd-arm@freebsd.org> Subject: Re: MMC/SDIO stack under CAM Message-ID: <CAJ-VmonPkdVVq7nC3FdopcgzmSTsj3gTO=Cghx-62XS5K25YQg@mail.gmail.com> In-Reply-To: <53120EE8.1080600@bakulin.de> References: <20140216111153.GA74858@olymp.kibab.com> <5C2CF572-360D-4CA0-81C7-18A5C455AED5@bsdimp.com> <20140224142642.GA32538@olymp.kibab.com> <CAJ-VmomNzCMc1T=0jAnyd_uGXbvgeTzZTtmhUPSvZ0DKUEjtKg@mail.gmail.com> <53120EE8.1080600@bakulin.de>
next in thread | previous in thread | raw e-mail | index | archive | help
On 1 March 2014 08:46, Ilya Bakulin <ilya@bakulin.de> wrote: > Hi Adrian, > > On 24.02.14, 16:59, Adrian Chadd wrote: >> hi, >> >> Let me just reiterate some .. well, experience doing this stuff at QCA. >> >> You really, absolutely don't want too much overhead in the MMC/SDIO >> path between whatever is issuing things and the network driver. >> >> There was significant performance work done at QCA on a local MMC/SDIO >> driver and bus to get extremely low latency and CPU utilisation when >> pushing around small transactions. The current CAM locking model is >> not geared towards getting to high transaction rates. > > So here you mean some work done on Linux MMC/SDIO stack by QCA > which made it far better than current Linux MMC stack in terms of > high SDIO I/O rates? Yup. The stock MMC stack/driver in Linux wasn't "fast" enough at small transactions to sustain the wifi speeds customers required. >> >> You may think this is a very architecturally pretty solution and it >> indeed may be. But if it doesn't perform as well as the existing local >> hacks that vendors have done, no company deploying this hardware is >> going to want to use it. They'll end up realising there's this massive >> CAM storage layer in between and either have to sit down to rip it up >> and replace it with something lightweight, or they'll say "screw it" >> and go back to the vendor supplied hacked up Linux solution. > > I think that if the "architecturally pretty solution" behaves worse than > some ugly hacks, then it may be not so pretty or the architecture is > just broken > by design. > >> So I highly recommend you profile things - and profile things with >> lots of small transactions. If the CAM overhead is more than a tiny, >> tiny fraction of CPU at 25,000 pps, your solution won't scale. :-) > > I don't really know what to compare with. For MMC/SD cards it is pretty > obvious, but then these cards will be likely the bottleneck, not the stack. > And the only goal would be to not make the stack slower than it is now. > But, as ATA devices are much faster than MMC/SD, I don't think this will > be a problem. > > For SDIO things are different. But we don't have any drivers (yet), except > mv_sdiowl that I'm writing, to test on. So I have to bring the SDIO > stack on CAM, > than bring mv_sdiowl to the state when it can actually transmit the > data, and then > compare performance with the vendor-supplied Linux driver. > We'll see then if there is a room for improvement... That sounds like a plan. Just note that although storage looks like it's doing much more throughput, the IO size also matters. As I said above, it's not uncommon to have > 1000 receive frames a second on 802.11n; and that can peak much higher than that. That's not the kind of IO rate you see on SD cards. :-) -a
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAJ-VmonPkdVVq7nC3FdopcgzmSTsj3gTO=Cghx-62XS5K25YQg>