During random reads, disk activity is usually low, as measured by io reads/sec. You are spending all your time seeking.
If it truly was totally cached, you are now testing the RPC cycle speed, which is currently around 1000 ops/sec/thread. If you retrieve 1000 rows at a time, the speed is good, but if you retrieve 1 per second... well I think you get the picture. -ryan On Fri, Dec 11, 2009 at 5:38 PM, stack <[email protected]> wrote: > No disc activity across all of the cluster -- i.e. we the disc activity is > not happening elsewhere on the cluster -- and the log of cache activity > shows that cache is loaded (and the amount of data loaded all fits in RAM?)? > > St.Ack > > On Fri, Dec 11, 2009 at 12:22 PM, ruslan usifov > <[email protected]>wrote: > >> Yes this all obvious, but i have no any disc activity (all data was cached >> by Hbase) >> >> disk seek speed (typically 5-9ms) and how much locality your random reads >> > >> > >> >
