[rrd-developers] rrdtool 1.3 cairo switch is done
Dylan Vanderhoof
DylanV at semaphore.com
Fri Jun 15 20:40:58 CEST 2007
I'm not sold that it's the cairo stuff causing the performance
degredation. fetch shouldn't have anything to do with the graphics lib,
and its running at less than half the speed of the 1.2.23 + fadvise
rrdtool.
Graphing is ~1/3 the speed of 1.2.23, but its hard to say how much of
that is cairo and how much is rrdtool itself.
(FWIW, I get nearly identical results without O_DIRECT)
-D
> -----Original Message-----
> From: Matthew Chambers [mailto:matthew.chambers at vanderbilt.edu]
> Sent: Friday, June 15, 2007 11:29 AM
> To: Dylan Vanderhoof; rrd-developers at lists.oetiker.ch
> Subject: Re: [rrd-developers] rrdtool 1.3 cairo switch is done
>
>
> > On closer look, this seems to be rrdtool and not the new graph
> > libraries.
> >
> > denver ~ # time ./rrdtool fetch test.rrd AVERAGE
> > real 0m0.041s
> > user 0m0.000s
> > sys 0m0.030s
> >
> > denver ~ # time rrdtool fetch test2.rrd AVERAGE
> > real 0m0.017s
> > user 0m0.010s
> > sys 0m0.000s
> > denver ~ #
> >
> > (rrd files above are identical, just using different inodes
> to prevent
> > caching)
> >
> > -D
> >
> > > -----Original Message-----
> > > From: Dylan Vanderhoof
> > > Sent: Friday, June 15, 2007 10:50 AM
> > > To: rrd-developers at lists.oetiker.ch
> > > Subject: Re: [rrd-developers] rrdtool 1.3 cairo switch is done
> > >
> > >
> > > Tried this on my test server. Generated 515 graphs of
> ~30 days each.
> > > Trunk functions fine and the graphs look good, but its
> substantially
> > > slower than 1.2.23 w/ the fadvise patch. (trunk compiled
> > > with mmap and
> > > directio)
>
> This is interesting stuff. I am interested in how much computation is
> necessary for Cairo/Pango vs. libart? What is the hypothesis
> and what would
> be a good benchmark for testing this change?
>
> -Matt
>
> _______________________________________________
> rrd-developers mailing list
> rrd-developers at lists.oetiker.ch
> https://lists.oetiker.ch/cgi-bin/listinfo/rrd-developers
>
More information about the rrd-developers
mailing list