Re: [Bigdata-developers] BSBM 100M load performance

[Bryan T. <br...@sy...>]

That is at 98% disk utilization, so clearly the bottleneck (outside of Q3) is the disk. Bryan 

> -----Original Message-----
> From: Bryan Thompson [mailto:br...@sy...] 
> Sent: Friday, July 30, 2010 2:53 PM
> To: Bigdata Developers
> Subject: Re: [Bigdata-developers] BSBM 100M load performance
> 
> The numbers below were with 4 clients.  Running with 8 
> concurrent clients, it rings in 4150 QMpH. 
> 
> Again, these are not official scores since we are not running Q3.
> 
> Bryan
> 
> > -----Original Message-----
> > From: Bryan Thompson [mailto:br...@sy...]
> > Sent: Friday, July 30, 2010 2:43 PM
> > To: Bigdata Developers
> > Subject: Re: [Bigdata-developers] BSBM 100M load performance
> > 
> > Just to follow up on query performance, I am ringing in 2408 QMpH 
> > (query mixes per hour) for the reduced BSBM query mix with Q3 
> > excluded.  This is not an "official" score since you can't 
> modify the 
> > benchmark, and I think that includes chosing to exclude a query.  
> > However, if it were an official result, it would be the 
> "high score" 
> > for BSBM.  This is running on my Centos 4 core AMD 16G workstation 
> > with striped RAID on 15k SAS drives.  We score 7x that on SSD.
> > 
> > Q3 involves an optional join (the entire join is optional).  
> > We currently delegate that to Sesame and the performance for
> > Q3 blows -- it just drags everything down.  Once we execute that 
> > natively we can (finally) publish a real BSBM score.
> > 
> > Bryan
> > 
> > > -----Original Message-----
> > > From: Bryan Thompson [mailto:br...@sy...]
> > > Sent: Friday, July 30, 2010 2:03 PM
> > > To: Bigdata Developers
> > > Subject: [Bigdata-developers] BSBM 100M load performance
> > > 
> > > Just to let everyone know, the load rate on the BSBM 100M
> > dataset on a
> > > WORM Journal has gone up significantly (~2x) with the 
> inline of XSD 
> > > numerics into the statement indices and the size on the disk has 
> > > decreased by nearly 25%.  Query performance is roughly the
> > same, with
> > > the interesting exception that we can double the performance when 
> > > changing the seed - presumably because there is less data
> > on the disk
> > > and therefore there is now a buffering effect when changing seeds 
> > > between runs.
> > > 
> > > Thanks,
> > > Bryan
> > > --------------------------------------------------------------
> > > ----------------
> > > The Palm PDK Hot Apps Program offers developers who use 
> the Plug-In 
> > > Development Kit to bring their C/C++ apps to Palm for a 
> share of $1 
> > > Million in cash or HP Products. Visit us here for more details:
> > > http://p.sf.net/sfu/dev2dev-palm
> > > _______________________________________________
> > > Bigdata-developers mailing list
> > > Big...@li...
> > > https://lists.sourceforge.net/lists/listinfo/bigdata-developers
> > > 
> > --------------------------------------------------------------
> > ----------------
> > The Palm PDK Hot Apps Program offers developers who use the Plug-In 
> > Development Kit to bring their C/C++ apps to Palm for a share of $1 
> > Million in cash or HP Products. Visit us here for more details:
> > http://p.sf.net/sfu/dev2dev-palm
> > _______________________________________________
> > Bigdata-developers mailing list
> > Big...@li...
> > https://lists.sourceforge.net/lists/listinfo/bigdata-developers
> > 
> --------------------------------------------------------------
> ----------------
> The Palm PDK Hot Apps Program offers developers who use the 
> Plug-In Development Kit to bring their C/C++ apps to Palm for 
> a share of $1 Million in cash or HP Products. Visit us here 
> for more details:
> http://p.sf.net/sfu/dev2dev-palm
> _______________________________________________
> Bigdata-developers mailing list
> Big...@li...
> https://lists.sourceforge.net/lists/listinfo/bigdata-developers
>