Harddisk benchmarks with ZCAV and gnuplot
I was browsing the campzone forums and came across a topic where everyone was showing their HD Tune graphs. Since I haven’t done any benchmarks on my new server yet, I became interested. Unfortunately HD Tune is only for windows so I fired up google to look for alternatives for BSD. I came across a site that used the output of ZCAV (a tool from the bonnie++ package) with gnuplot to create some cool graphs. I decided to try it out and installed gnuplot and bonnie++. After installing, I ran the zcav command for both my array’s:
zcav -c3 -laacd0 -f/dev/aacd0
zcav -c3 -laacd1 -f/dev/aacd1
This ran for a while and after that, the two logfiles aacd0 and aacd1 were created.
The -c3 specifies that zcav should run the test 3 times, this results in nicer graphs.
ZCAV reads the data from the harddisk and divides this up in zones, you then get the per-zone speed. Zones on the outside of the disk platter are faster than zones on the inside.
Then with the following gnuplot script I generated a graph:
unset autoscale x
set autoscale xmax
unset autoscale y
set autoscale ymax
set xlabel "Position MB"
set ylabel "KB/s"
set key right bottom
plot "aacd0" title "4x ST373207LC RAID10", "aacd1" title "2x ST373207LC RAID1"
set terminal png
set output "disks.png"
replot
The red array contains my /, the green array is for /home
The horizontal line shows that there is a bottleneck somewhere, the disks can go faster, but something is capping it at 110MB/s. Anyway, it’s a pretty good speed i’m getting there. These disks are all connected to one channel of the servers Adaptec AIC-7902 combined with the AOC-LPZCR2. The storage array is connected to the other channel.
After I finished the zcav run for my storage array, which I did with -c1 because of the amount of data, I made a graph of that too:
The Seagate’s are performing great. Especially the 250GB versions who stay at 55MB/s on the entire platter. Then the two Hitachi Deskstars, which are nicknamed Deathstars at my work, it seems these two are going to die too, judging by the speed drop on a large part of the disk. And then the Maxtors, which are just bad overall.
Comments
-
So should this be fast?










