On Thu, Aug 3, 2017 at 11:07 PM, Johannes Thumshirn <jthumshirn@xxxxxxx> wrote: > On Thu, Aug 03, 2017 at 09:15:12AM -0700, Dan Williams wrote: >> > I can put in a tunable knob like Johannes suggested and leave the >> > default for that to what it is currently. And we can adjust as necessary >> > as we do more testing. >> >> That's the problem, we need a multi-dimensional knob. It's not just >> transfer-size that effects total delivered bandwidth. > > If Dan opposes it so much, I'm OK with no tunable. I'm just expressing my > concern that we get reports like this [1] one here as a result. Sorry if I came off harsh. In that report from Mel he's showing a slow down in a representative workload not just a canned bandwidth benchmark. I'm worried that tuning for a single dimension on an fio run doesn't actually improve throughput delivered to an application. -- To unsubscribe from this list: send the line "unsubscribe dmaengine" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
