Hello David, On Thu, 28 Mar 2019 at 23:07, David Byte <dbyte@xxxxxxxx> wrote: > > I am trying to figure out how I would do a test where I have multiple ranges of block sizes with a percentage distribution assigned to each range. > > Something like this but works and reports the io data for each range: fio --bssplit=1k-8k/10:8k-128k/50 --name=test --rw=randrw --time_based=1 --runtime=30 --size=1G --filename=/tmp/test. > > Clearly using a jobfile with named sections is a start, but I'm not sure how to control the distribution of I/O between the sections? > > Thoughts or better yet an example? I can give a thought... If you are trying to distribute I/O *between two sections* you can look towards flow (https://fio.readthedocs.io/en/latest/fio_doc.html#cmdoption-arg-flow ). If you need it between N>2 sections (and not just pairs) I don't think flow is going to be good enough and I don't have any suggestions (yes multiple sections can work with the same counter but there are only two sides to it's more like a tug of war). What would be excellent is if someone were to add a feature that allows flow to work in a "shares" mode. Every active job could say how many shares they have and then each time they do I/O a share is used and you can only do I/O while your shares are above 0. A job's shares would only be replenished after every job's shares have become 0. I suppose so long as the total shares added up to 100 people could even see them percentages (but that may be a bit bursty and maybe you want to try and order the jobs' "goes" in a way that everyone's shares to drop to 0 at roughly the same time). There's also issues around jobs that finish early while others are ongoing and jobs etc. but I still think it could be made to work. -- Sitsofe | http://sucs.org/~sits/
