Hi, On 20 April 2017 at 23:12, Sitsofe Wheeler <sitsofe@xxxxxxxxx> wrote: > > On 20 April 2017 at 15:03, kjaszczerski <kjaszczerski@xxxxxxxxx> wrote: >> >> I'm trying to run FIO on mmap engine on 4k socket and 4k channel. I'd like >> to use wrokload with blocksize parameter value less than 4k, is it possible >> ? > > You didn't give an example jobfile, fio version, operating system > (which are really necessary to answer these types of questions) so the > following is going to be a lot of speculation. > > I don't know if you also filed https://github.com/axboe/fio/issues/339 > but there's a partial answer there (using direct=1 forces alignment > and size constraints). There's a follow up point which is the message > stems from a desire to catch msync related errors earlier: > https://github.com/axboe/fio/commit/913ea0db12596b735a030a1836bd913497b7120a > . I suppose a filesystem will never have a blocksize smaller than a > page and alignment constraints mean an msync must always start page > aligned (http://man7.org/linux/man-pages/man2/msync.2.html ) so if you > have any type of fsync taking place that will trigger that check. The above is slightly incorrect because the requirement for page alignment on the address passed to msync() must be obeyed regardless of your filesystem/device. Since the fio mmap ioengine approximates direct I/O by using msync() (see https://github.com/axboe/fio/blob/fio-2.19/engines/mmap.c#L189 ) even a job using that ioengine which only asks for direct=1 will have to obey the "all offsets must be page aligned" requirement. In order to satisfy that requirement, when using direct=1 with the mmap ioengine the blocksize must always be a (non-zero) multiple of the page size and if your page size is 4096 this means you can't use a block size less than that. -- Sitsofe | http://sucs.org/~sits/ -- To unsubscribe from this list: send the line "unsubscribe fio" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
