> -----Original Message----- > From: Alireza Haghdoost [mailto:haghdoost@xxxxxxxxx] > Sent: Thursday, 08 January, 2015 1:01 PM > To: Elliott, Robert (Server Storage) > Cc: Jens Axboe; fio@xxxxxxxxxxxxxxx > Subject: Re: Non-uniform randomness with drifting > > On Thu, Jan 8, 2015 at 10:59 AM, Elliott, Robert (Server Storage) > <Elliott@xxxxxx> wrote: > > The new random_drift feature changes spatial locality over time (where > > time means % of data that has been accessed), so it provides some > > correlation control, but doesn't provide full temporal locality > > control. > > Just to clarify, the temporal locality control that you are referring > to is more like burst control. You are not interested in a fixed IO > generation rate and interested to have dynamic IO rate changes. Is > that correct ? Yes. That kind of temporal locality is necessary for write buffer/caches that sit between a fast interface and a slow interface, where the buffer is continuously destaged. Assuming the long-term input rate matches the output/destage rate (otherwise the buffer will inevitably become full), the required buffer size to maintain maximum bandwidth depends on the length of the bursts between the idles. A simple fixed rate generator has uses too, but it often leads to overloading the system or settling at less than the maximum possible rate. That's how the SPC-1 benchmarks are designed; find the heaviest workload for which the system maintains the targeted latency (fio's latency_target= option). -- 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
