[Linux-cluster] GFS on 2.6.8.1 more simple performance numbers

[Daniel McNeil]

On Mon, 2004-10-18 at 17:39, Daniel McNeil wrote:
> On Mon, 2004-10-18 at 14:38, Kevin Anderson wrote:
> [snip]
> > 
> > Can you rerun the ext3 numbers to the same storage as GFS?  Would also
> > be interesting to run from two nodes simultaneously to separate ext3
> > filesystems on the storage as well?  Doing dd's of the logical
> > partitions would also be interesting to see what bandwidth the storage
> > is capable of providing, first single node then simultaneous from both
> > nodes.
> > 
> 
> Partial ext3 results:
> 
> ext3 on 5 disk stripe
> =====================
> time tar xf /Views/linux-2.6.8.1.tar; time sync
> real    0m5.221s user    0m0.442s sys     0m4.238s
> real    0m8.214s user    0m0.000s sys     0m0.390s
> 
> (1st time)time du -s linux-2.6.8.1/
> 226916  linux-2.6.8.1/
> real    0m2.888s user    0m0.035s sys     0m0.288s
> 
> time rm -rf linux-2.6.8.1/
> real    0m2.131s user    0m0.022s sys     0m0.821s
> 
> data=writeback
> --------------
> time tar xf /Views/linux-2.6.8.1.tar; time sync
> real    0m5.443s user    0m0.439s sys     0m4.056s
> real    0m8.774s user    0m0.001s sys     0m0.388s
> 
> time rm -rf linux-2.6.8.1/; time sync
> real    0m1.127s user    0m0.035s sys     0m1.012s
> 
> time du -s linux-2.6.8.1/
> 226916  linux-2.6.8.1/
> real    0m2.741s user    0m0.058s sys     0m0.281s
> 
> data=log
> ========
> time rm -rf linux-2.6.8.1/; time sync
> real    0m2.106s user    0m0.024s sys     0m0.824s
> 
> time du -s linux-2.6.8.1/
> 226916  linux-2.6.8.1/
> real    0m2.738s user    0m0.025s sys     0m0.312s
> 
> time tar xf /Views/linux-2.6.8.1.tar; time sync
> real    0m22.784s user    0m0.445s sys     0m4.639s
> real    0m1.234s user    0m0.001s sys     0m0.248s

I partitioned my 5 disk stripe into 2 partitions and
ran the tar/du/rm test from 2 ext3 file systems from
2 nodes simultaneously.

2 ext3 file systems on separate partitions of a 5-disk stripe
=============================================================
                                                                                
time tar xf /Views/linux-2.6.8.1.tar; time sync
real    0m6.574s user    0m0.498s sys     0m4.097s
real    0m20.915s user    0m0.000s sys     0m0.480s
                                                                                
cl030]# time tar xf /Views/linux-2.6.8.1.tar; time sync
real    0m6.324s user    0m0.460s sys     0m4.179s
real    0m21.434s user    0m0.001s sys     0m0.459s
                                                                                
time du -s linux-2.6.8.1/
226916  linux-2.6.8.1/
real    0m3.207s user    0m0.054s sys     0m0.264s
                                                                                
cl031]# time du -s linux-2.6.8.1/
226916  linux-2.6.8.1/
real    0m2.890s user    0m0.037s sys     0m0.291s
                                                                                
time rm -rf linux-2.6.8.1/; time sync
real    0m2.265s user    0m0.029s sys     0m0.809s
real    0m0.194s user    0m0.000s sys     0m0.005s
                                                                                
time rm -rf linux-2.6.8.1/; time sync
real    0m2.441s user    0m0.025s sys     0m0.835s
real    0m0.176s user    0m0.001s sys     0m0.003s

It looks like the tar time from 2 ext3 file systems to 
the same 5 disk stripe is taking twice as long.  The
du and rm times are almost the same as 1 ext3 file system
on the 5 disk stripe alone.

Daniel