[Pvfs2-users] Observations related to mail of 02/04/2008 Expected/ToExpect vs. real performance ??

[belcampo]

First of all a VERY BIG THANK YOU.

I'm experimenting with pvfs2 and are very impressed of what can be done 
with it, installation/performance wise. Even I can do it. :-)

A) 3 machines are meta-server io-server and client
copy from pvfs-fs to local-fs
0.82user 36.30system 12:31.70elapsed 4%CPU (0avgtext+0avgdata 
0maxresident)k0inputs+0outputs (18major+710minor)pagefaults 0swaps

B) from pvfs-fs to same pvfs-fs with other file-name
1.13user 36.09system 18:29.67elapsed 3%CPU (0avgtext+0avgdata 
0maxresident)k0inputs+0outputs (1major+205minor)pagefaults 0swaps

C) 1st 'external-client' not being part as io-server or whatever
copy from pvfs-fs to local-fs
0.00user 26.15system 2:14.80elapsed 19%CPU (0avgtext+0avgdata 
0maxresident)k0inputs+0outputs (0major+1228minor)pagefaults 0swaps

D) 1st and 2nd 'external-client' not being part as io-server or whatever
simultaneous copy same-file from pvfs-fs to their own local-fs
0.00user 18.16system 3:30.86elapsed 8%CPU (0avgtext+0avgdata 
0maxresident)k0inputs+0outputs (2major+1225minor)pagefaults 0swap

E) 1st and 2nd 'external-client' not being part as io-server or whatever
simultaneous copy same-file from local to pvfs-fs with different name
0.00user 22.58system 3:19.63elapsed 11%CPU (0avgtext+0avgdata 
0maxresident)k0inputs+0outputs (3major+1224minor)pagefaults 0swaps

F) 1st 'external-client' not being part as io-server or whatever
copy from pvfs-fs to pvfs-fs with different name
0.00user 18.16system 3:30.86elapsed 8%CPU (0avgtext+0avgdata 
0maxresident)k0inputs+0outputs (2major+1225minor)pagefaults 0swaps

Physically the result of B) and F) is the same, a copy is made on the 
pvfs-fs. The needed time however is the difference between day and night.

Throughput in these cases varies between 7.5MB/s and > 100MB/s, pretty 
close to what 1Gb network can do.

King regards

Henk Schoneveld