It seems I jumped the gun too early here. I've been running tests for the last 24 hours and I just had one squirt across multiple clients rendering the last tick unusable data. Testing this at my scale with another output engine requires me to rewrite my parsers and I might end up doing that but I'd hoped there was a more simplistic workaround for this problem. Thanks! Michael On Fri, Oct 24, 2014 at 12:22 PM, Michael Mattsson <michael.mattsson@xxxxxxxxx> wrote: > Hey, > I took another stab at reshuffling output control and I've found a > combo that works very consistently. By dropping --output and piping to > tee instead I get an incredible high accuracy in my 'runtime' output. > It only differs 1-2ms between a sample of ~500 runs. Before this was > literally all over the place (+/- <1000ms). > > Could it be that output redirection is blocked when writing with with > fio and asynchronously when piping to tee? Output is being written to > an NFS mount which is subject to several layers of virtualization and > may cause intermittent latency spikes and deliver my very inconsistent > results. Thoughts? > > Anyhow, thank you for your work here and I owe you lunch! > > Regards > Michael > > On Fri, Oct 24, 2014 at 8:50 AM, Jens Axboe <axboe@xxxxxxxxx> wrote: >> On 2014-10-24 09:44, Michael Mattsson wrote: >>> >>> Hey, >>> Is there anything I can do to help you look at this issue? >> >> >> Does it only happen for minimal output, or on "normal" or json output too? >> >> -- >> Jens Axboe >> -- 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
