Humor me and test throughput with 100 threads first. :)
On Wed, Dec 16, 2009 at 9:30 AM, Richard Grossman <[email protected]> wrote: > I'm not using 50 thread but make it with 4 thread. > I give 2 thread by server ip. So I insert using 2 thread on 1 machine and > with 2 other on the second machine > > I need to add a lot of thread to be able to insert this data quickly enough. > > but for you it's logical this behavior ? > > > > On Wed, Dec 16, 2009 at 4:45 PM, Jonathan Ellis <[email protected]> wrote: >> >> Sounds like you are using a single thread, so the increased latency is >> artificially reducing your numbers. Add more threads (stress.py uses >> 50 by default) to get more throughput. (Also true even for a single >> node, but more noticable when you add network overhead to the >> cluster.) >> >> -Jonathan >> >> On Wed, Dec 16, 2009 at 8:06 AM, Richard Grossman <[email protected]> >> wrote: >> > Hi >> > >> > I think someone ask already similar but can't find where. >> > >> > On 1 machine standalone I insert data I get ~850 rows / second >> > On another machine I make exactly the same operation I get ~900/1000 >> > rows / >> > second >> > >> > Now I remove all the data from the 2 machines. Take exactly the same >> > storage-conf.xml but just add seed in both file nothing else. >> > Make the insert I get ~90 rows / second. >> > >> > Someone have an idea why the performance could fall sharply like this. >> > Or >> > simply give a hint what or where to check why it's happend >> > I've already checked network problem the 2 machines are identical. >> > >> > Thanks. >> > >> > >> > > >
