Hi, Let's keep the list CC'd - someone else may have the same question and find this thread. On Sat, 8 Jun 2019 at 14:51, Kurt Dorsey <batlin@xxxxxxxxx> wrote: > > Sitsofe, > I need to do this single job, Looking at the *roundrobin* option for > --filename, random seems to generate a new offset for every file yet > for sequential it only grabs a new offset after it cycles through all > files. Is there a way to add an option for it to act like random > (grab new offset for every file? Or if someone could point me at the > place in the code I could create my own internal patch for this > scenario? I don't think the code currently supports this in a generative manner as it would need to track the starting offset on a per file basis. I'd start looking at get_io_u() inside io_u.c (i.e. https://github.com/axboe/fio/blob/bc596cbcdbb58b81da53a29acf1370d8a7e94429/io_u.c#L1301 ) for the place to modify. Also bear in mind that while fio already knows how do gappy I/O via rw=write:<nr> (e.g. if my LBAs are 512 bytes big and I'm happy to write 4 of them in one go by using bs=2k then leave a 2k gap I can usebs=2k rw=write:2k) if you want to use a blocksize of a size smaller than a land you will likely need to write more code to support that too. Have you thought about faking up the sequence you want via replay (https://fio.readthedocs.io/en/latest/fio_doc.html#i-o-replay or https://fio.readthedocs.io/en/latest/fio_doc.html#trace-file-format-v2)? -- Sitsofe | http://sucs.org/~sits/
