pveperf fsync performance slower with raid10 than raid1?

[VinceO]

I've got 2 sets of Intel 4510 3.84TB drives, a 2-drive zfs raid1 and a 4-drive zfs raid10. Running pveperf i get slower fsync results with the raid10. Is that expected? both pools are in the same server on the same controller and all drives are identical.


root@pve2:~# pveperf /raid1-ssd-pool/
CPU BOGOMIPS: 259276.56
REGEX/SECOND: 1757737
HD SIZE: 1026.72 GB (raid1-ssd-pool)
FSYNCS/SECOND: 4056.35
DNS EXT: 24.71 ms
DNS INT: 0.94 ms (.com)

root@pve2:~# pveperf /raid10-ssd-pool/
CPU BOGOMIPS: 259276.56
REGEX/SECOND: 1783664
HD SIZE: 820.30 GB (raid10-ssd-pool)
FSYNCS/SECOND: 3791.16
DNS EXT: 35.67 ms
DNS INT: 0.87 ms (.com)

 

[czechsys]

2x 3.84T in zfs r1 = 1TB? what?
HD SIZE: 1026.72 GB (raid1-ssd-pool)

4x 3.84T in zfs r10 = 820GB? wtf?
HD SIZE: 820.30 GB (raid10-ssd-pool)

 

[spirit]

pveperf fsync do a simple test writing in 1 file in loop

           my $m = $count % 300;

            my $filename = "$dir/tf_$m.dat";

            open (TMP, ">$filename") || die "open failed: $!\n";
 
            print TMP $data;

            PVE::Tools::fsync(fileno(\*TMP));

            close (TMP);

            $elapsed = tv_interval ($starttime);

            last if $elapsed > 3;
        }

So It's possible that it's writing only on a couple of raid1 disk , in the raid10 array.

you should try to do a test with a true benchmark tool like fio.

 

[fabian]

like @spirit said - pveperf is not really a benchmark tool it just gives you a quick ballpark estimate of system performance