On 9/5/24 12:05 PM, Linus Walleij wrote: > On Thu, Sep 5, 2024 at 4:58?PM Jens Axboe <axboe@xxxxxxxxx> wrote: >> On 9/5/24 7:03 AM, Linus Walleij wrote: > >>> Which production? For singlequeue devices it is pretty widespread. >> >> We tried it at one point internally at Meta, and it was not pretty. > > I didn't know you used any singlequeue devices. > > If you used it on multiqueue devices, well that can't be recommended. Of course we have single queue devices, anything that isn't nvme is basically single queue. >>> Maybe we should propose these rules to the main udev repository >>> so that they also go into Debian and we get even wider use? >> >> I know you like to push for it to be the default, and I always push back >> because I don't think it's stable enough for that, and now we have the >> added complication that it hasn't been maintained for quite a while. >> So no, I don't think so. > > The reason I like it personally is that it has actually saved me from > crashing my machine by preserving interactivity on a (single queue) > device: > https://people.kernel.org/linusw/bfq-saved-me-from-thrashing I'd consider that largely anecdotal at this point. > For Androids and chromebooks it keeps the device interactive > during heavy disk (eMMC) activity, such as when Android > updates a pile of apps (.apk files). I'm somewhat dubious that that problem could not be solved in a much simpler way than what is BFQ. -- Jens Axboe
