Lars, If increasing number of worker threads and using connection pooling did not improve performance, then maybe we're looking at some kind of environmental or setting issue.
But just in case -- can you post snippets of the code that's setting up the java riak client and issuing the writes? On Fri, Feb 22, 2013 at 5:09 PM, Lars J. Nilsson <[email protected]> wrote: > Sorry, yes. I've used anything from single threads to one thread per > "actor". Including anything from one connection to one per process, and > pooling, etc. > > /Lars J. Nilsson > [sent from my mobile] > > Den 22 feb 2013 19:49 skrev "Sean Cribbs" <[email protected]>: > >> Use multiple threads in your Java program to parallelize the writes. >> >> On Fri, Feb 22, 2013 at 9:40 AM, Lars J. Nilsson >> <[email protected]> wrote: >> > Hi Sean, >> > >> > Thanks for pipjng in. And how do I do that? >> > >> > Cheers >> > /Lars J. Nilsson >> > Executive VP, Cubeia Ltd >> > +46 (0) 704 / 10 69 53 >> > >> > >> > On 22 February 2013 16:28, Sean Cribbs <[email protected]> wrote: >> >> >> >> Given iostat says ~10% utilization, I would suggest increasing the >> >> concurrency of your writes. It's obviously not using up the capacity >> >> of the disk, your bottleneck is elsewhere. 2MB/sec is low for any >> >> commodity hardware. >> >> >> >> On Fri, Feb 22, 2013 at 9:21 AM, Lars J. Nilsson >> >> <[email protected]> wrote: >> >> > Hi all, >> >> > >> >> > I'm micro-testing a simple update pattern (key + short json document) >> >> > on >> >> > my >> >> > local workstation (8 core intel CPU, 8 gig of RAM, 10K disk) to get >> >> > baseline >> >> > before moving to cluster on AWS. With Riak I get about 60 mods/s >> >> > before >> >> > maxing out (as compared to approx 3000 mods/s on a vanilla MySQL >> >> > using >> >> > exactly the same code). >> >> > >> >> > It feels like I must be doing something hideously wrong. (For the >> >> > record, >> >> > I'm a decent Java programmer but a riak/erlang n00b). >> >> > >> >> > Riak is installed via a vanilla Ununtu 12.04 package as described in >> >> > the >> >> > docs. The test ran on the same disk as the mentioned MySQL test. The >> >> > same >> >> > Java code was doing both tests (just a DAO swapped). >> >> > >> >> > The bucket was using n_val = 1. >> >> > >> >> > The problem seems to be in high disk-writing, despite the low load. >> >> > Here's >> >> > what iostat says during the run: >> >> > >> >> > [...] >> >> > wrqm/s : 434.50 >> >> > w/s: 182.90 >> >> > wkB/s: 2618.00 >> >> > avgrq-sz: 28.63 >> >> > [...] >> >> > %util: 10.60 >> >> > >> >> > Which seems to be too much considering the small load (the stored >> >> > JSON >> >> > blobs >> >> > are roughly 1K each). >> >> > >> >> > I'm using the official PB Java client. >> >> > >> >> > This is an evaluation for a customer. Any hint of what I'm missing or >> >> > what >> >> > the problem could be would be appreciated. >> >> > >> >> > Cheers >> >> > /Lars J. Nilsson >> >> > Executive VP, Cubeia Ltd >> >> > +46 (0) 704 / 10 69 53 >> >> > >> >> > _______________________________________________ >> >> > riak-users mailing list >> >> > [email protected] >> >> > http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com >> >> > >> >> >> >> >> >> >> >> -- >> >> Sean Cribbs <[email protected]> >> >> Software Engineer >> >> Basho Technologies, Inc. >> >> http://basho.com/ >> > >> > >> >> >> >> -- >> Sean Cribbs <[email protected]> >> Software Engineer >> Basho Technologies, Inc. >> http://basho.com/ > > > _______________________________________________ > riak-users mailing list > [email protected] > http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com > _______________________________________________ riak-users mailing list [email protected] http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
