That 100% utilization turned out to be something else messing up my run. However, I did find what was holding up the client. By default the benchmark only allows 100 outstanding requests. You can fix this by setting -Dnpar=XXXX on the command line.
-Ivan On Fri, Apr 20, 2012 at 03:35:06PM +0200, Ivan Kelly wrote: > Thats great :) > > I still think the client is saturated though, so you should be able to > get higher than that. I'm trying to figure out where the saturation is > coming from. 1 core is at 100% utilization, while the rest are at 97% > idle. The bottleneck isn't immediately obvious though. I'll keep you > updated. > > Regards > Ivan > > > On Thu, Apr 19, 2012 at 03:50:37PM -0700, Aniruddha Laud wrote: > > I will run the bookie benchmark a few times again to make certain the > > numbers are correct :) It seems I had logging turned on for the client. I > > disabled it and I get a much higher tput now. Around 20k for 30 topics and > > message size 100. > > > > Regards, > > Aniruddha. > > > > On Thu, Apr 19, 2012 at 3:06 PM, Ivan Kelly <[email protected]> wrote: > > > > > 60k per bookie? What kind of disks are in these machines. That > > > actually seems far too high :) > > > > > > Have you checked that the problem isn't that the client is > > > saturating. What kind of processing power do the hubs have? Have you > > > configured the hubs different to how they are by default? > > > > > > I used the java client. > > > > > > -Ivan > > >
