Hi Frederick, Any chance of sharing your test harness? I think such a performance comparison tool would be useful, both for the metrics it provides and also to show how the same queries might be modeled/executed in Phoenix vs. raw HBase API. A very good example application for the communities.
Thanks, Nick On Wed, Mar 12, 2014 at 5:57 AM, Frederick Haebin Na <[email protected]>wrote: > Okay, the slow performance of the salted table was due to the small > connection pool size config in the Spring's JDBC, not because of the > Phoenix's default conf. > > Best, > Haebin > > > > 2014-03-12 12:06 GMT+09:00 Frederick Haebin Na <[email protected]>: > > Hello all, >> >> I am doing a performance benchmark of Phoenix. >> For now, it just an insertion scenario, yet it turned out to be somewhat >> strange. >> >> I have 3 small node cluster for HBase with Phoenix. >> >> Insertion of 5K document with long key. >> 1. Vanilla HBase: 720 TPS (115 ms Mean Transaction Time) >> 2. Phoenix: 413 TPS (238 ms MTT) >> 3. Phoenix Salted (12 bucket): 195 TPS (514 ms MTT) >> >> Is the result plausible to you? >> Or should I tune something for phoenix? (such as client side thread pool >> size or so) >> >> Best, >> Haebin >> > >
