Admin please fix this thread

[andydills]

I'm trying to determine the best way to benchmark the I/O of my proxmox servers, because let's face it...I/O is the bottleneck until you can justify the move to a real iscsi implementation with racks of spindles at your disposal.

I have two servers that are doing drbd, RAID 10 with 3.6 TB usable on an Adaptec 6405 with BBU protected write caching enabled:

root@proxmox1:~# arcconf  GETCONFIG 1 | grep cache
   Read-cache mode                          : Enabled
   Write-cache mode                         : Enabled (write-back)
   Write-cache setting                      : Enabled (write-back)

My system drive, which is a simple mdadm RAID-1, scores as this:

# pveperf /
CPU BOGOMIPS:      95996.76
REGEX/SECOND:      863662
HD SIZE:           57.09 GB (/dev/mapper/pve-root)
BUFFERED READS:    114.24 MB/sec
AVERAGE SEEK TIME: 10.66 ms
FSYNCS/SECOND:     704.73
DNS EXT:           49.50 ms
DNS INT:           1.41 ms

So, ok, let me make a simple volume to test the performance of the hardware RAID:

# lvcreate -n test -L 10G data
# mkfs.ext3 /dev/data/test
# mount /dev/data/test /mnt
# pveperf /mnt

I was a little disappointed to see the results:

root@proxmox1:~# pveperf /mnt
CPU BOGOMIPS:      95996.76
REGEX/SECOND:      852961
HD SIZE:           9.84 GB (/dev/mapper/data-test)
BUFFERED READS:    430.93 MB/sec
AVERAGE SEEK TIME: 5.44 ms
FSYNCS/SECOND:     624.97
DNS EXT:           50.93 ms
DNS INT:           1.00 ms

From what I've seen, my fsyncs/sec are quite low.

Is drbd causing this perhaps? I haven't seen much in the way of benchmarks on drbd installations. My gut instinct is that something else is wrong...the scores are about what I'd expect if writeback cache wasn't enabled, but I set it in the BIOS, and I confirmed in the software, so I'm a bit stumped on what could be at fault here.


EDIT: This is quite interesting and perhaps relevant, I'm investigating this further now:

http://www.gossamer-threads.com/lists/drbd/users/23967