I'm doing some performance testing of glusterfs and I'm trying to figure out the best way to analyze the data generated. iozone looks like a pretty good tool and I should be able to script an ensemble that varies the performance settings, but actually analyzing the generated data is too big a job to do manually. I spent quite a while yesterday loading the results into Open Office and generating charts for four test runs. They look good, but it is far too large a job to do manually, an ensemble will probably consist of hundreds of runs. My maintenance windows are this week on Wednesday & Thursday and I have no idea when I will be allowed to take the cluster down again. Ideas? Please? -- Matt
Hi! 2008/12/29 Matt Lawrence <matt.lawrence at tamu.edu>:> I'm doing some performance testing of glusterfs and I'm trying to figure > out the best way to analyze the data generated. iozone looks like a > pretty good tool and I should be able to script an ensemble that varies > the performance settings, but actually analyzing the generated data is > too big a job to do manually. I spent quite a while yesterday loading > the results into Open Office and generating charts for four test runs. > They look good, but it is far too large a job to do manually, an > ensemble will probably consist of hundreds of runs. My maintenance > windows are this week on Wednesday & Thursday and I have no idea when I > will be allowed to take the cluster down again. > > Ideas? Please?gnuplot? (http://gnuplot.sourceforge.net/demo_4.2/) http://www.iozone.org/src/current/Gnuplot.txt Harald St?rzebecher
Hi! 2008/12/29 Matt Lawrence <matt.lawrence at tamu.edu>:> Harald St?rzebecher wrote: >> >> Hi! >> >> 2008/12/29 Matt Lawrence <matt.lawrence at tamu.edu>: >> >>> >>> I'm doing some performance testing of glusterfs and I'm trying to figure >>> out the best way to analyze the data generated. iozone looks like a >>> pretty good tool and I should be able to script an ensemble that varies >>> the performance settings, but actually analyzing the generated data is >>> too big a job to do manually. I spent quite a while yesterday loading >>> the results into Open Office and generating charts for four test runs. >>> They look good, but it is far too large a job to do manually, an >>> ensemble will probably consist of hundreds of runs. My maintenance >>> windows are this week on Wednesday & Thursday and I have no idea when I >>> will be allowed to take the cluster down again. >>> >>> Ideas? Please? >>> >> >> gnuplot? (http://gnuplot.sourceforge.net/demo_4.2/) >> >> http://www.iozone.org/src/current/Gnuplot.txt >> > > I can't find gen_graphs and gengnuplot.sh doesn't output a workable gnuplot > file. I'll have to keep looking. Thanks for the suggestions. I think > gnuplot is the way to go.http://www.iozone.org/src/current/report.pl worked for me after patching: diff report.pl report.pl.new 112c112 < set terminal png small picsize 900 700 ---> set terminal png small size 900 700125c125 < set terminal png small picsize 450 350 ---> set terminal png small size 450 350Harald St?rzebecher
On Mon, Dec 29, 2008 at 1:08 PM, Matt Lawrence <matt.lawrence at tamu.edu>wrote:> I'm doing some performance testing of glusterfs and I'm trying to figure > out the best way to analyze the data generated. iozone looks like a > pretty good tool and I should be able to script an ensemble that varies > the performance settings, but actually analyzing the generated data is > too big a job to do manually. I spent quite a while yesterday loading > the results into Open Office and generating charts for four test runs. > They look good, but it is far too large a job to do manually, an > ensemble will probably consist of hundreds of runs. My maintenance > windows are this week on Wednesday & Thursday and I have no idea when I > will be allowed to take the cluster down again. > > Ideas? Please? >The statistical package called "R" is great for this kinda thing. Sean -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20090105/becd87c8/attachment.html>