On 06/10/2010 11:39 AM, Andreas Kurz wrote:
On Thursday 10 June 2010 17:19:53 Michael Joyner wrote:
better, but still way below box2box bandwidth.... is fs type or fact
using lvm a factor? Here are my test results, will try w/o LVM next.
I can really recommend  "Part V. Optimizing DRBD performance" in the DRBD
Users Guide ... or invest in a "DRBD Healthcheck" offered by Linbit ;-)

Regards,
Andreas

I split it up into 3 volumes, to prevent split brain issue with LVM. :)
I retest with EXT4, same horrible results.
I then tested with XFS, much, much better results:
Migrated VMs over to it. They all seem to be running much better with XFS and not EXT4. As soon as I can get back into the COLO, I will reroute cabling to setup a RR bond between the two boxes directly to see if that helps.

Thanks for all the help.

=== *** === XFS Timing Test:

Large files test:

rsync box2box /var/tmp:

sent 6.96G bytes  received 164 bytes  47.16M bytes/sec

rsync local2/data w/ peer:

sent 6.96G bytes  received 164 bytes  41.53M bytes/sec

rsync local2/data no peer:

sent 6.96G bytes  received 164 bytes  114.99M bytes/sec

Iozone test (no peer):

real    0m4.983s
real    0m4.922s
real    0m4.920s
real    0m4.915s
real    0m4.917s

Iozone test (w/ peer);

real    0m29.644s
real    0m28.056s
real    0m27.634s
real    0m28.203s
real    0m28.209s

On 06/10/2010 04:05 AM, Andreas Kurz wrote:
.... LVM on your system supports barriers -->  drbd supports barriers -->
barriers are used to enforce write-after-write dependencies per default
... use drbd config-param 'no-disk-barrier'
=== new layout ==============================

using /dev/sdb as /dev/drbd0, non-spanned vg.

=== new drbd.conf ============================

global {
    usage-count yes;
}
common {
      protocol C;
      syncer { rate 700000K; }
      net {
          sndbuf-size 0;
          after-sb-0pri discard-least-changes;
          after-sb-1pri discard-secondary;
      }
      startup {
          degr-wfc-timeout 30000;
      }
      disk {
          no-disk-flushes;
          no-md-flushes;
          no-disk-barrier;
      }

}
resource sdb {
    on san-node-1.ewc.edu {
      device    /dev/drbd0;
      disk      /dev/sdb;
      address   192.168.75.201:7789;
      meta-disk internal;
    }
    on san-node-2.ewc.edu {
      device    /dev/drbd0;
      disk      /dev/sdb;
      address   192.168.75.202:7789;
      meta-disk internal;
    }
}

*********
=== *** mount /dev/vgdrbd0/nfs1 /data *** ===

=== iozone -a -e -g 4096 with peer down:
real    0m1.757s,  real    0m1.738s, real    0m1.753s, real
0m1.762s,  real    0m1.745s

=== with peer up:
real    0m6.739s, real    0m6.758s, real    0m6.619s, real    0m6.653s,
real    0m6.636s

=== big files test =============================

===  DRBD === rm -rfv /data/ISOS; rsync -a --verbose --human-readable
--progress ~mjoyner/ISOS /data/ISOS

building file list ...
8 files to consider
created directory /data/ISOS
ISOS/
ISOS/ISOS/
ISOS/ISOS/.lck-5a00b81000000000
            84 100%    0.00kB/s    0:00:00 (xfer#1, to-check=5/8)
ISOS/ISOS/.lck-6500b81000000000
            84 100%   82.03kB/s    0:00:00 (xfer#2, to-check=4/8)
ISOS/ISOS/.lck-7700b81000000000
            84 100%   82.03kB/s    0:00:00 (xfer#3, to-check=3/8)
ISOS/ISOS/SW_DVD5_NTRL_SQL_Svr_2008_SP1_English_X15-51857.ISO
       943.07M 100%  189.42MB/s    0:00:04 (xfer#4, to-check=2/8)
ISOS/ISOS/SW_DVD5_SQL_Svr_Enterprise_Edtn_2008_English_MLF_X14-89207.ISO
         3.26G 100%  189.81MB/s    0:00:16 (xfer#5, to-check=1/8)
ISOS/ISOS/SW_DVD5_Windows_Svr_2008w_SP2_English__x64_DC_EE_SE_X15-41371.ISO
         2.76G 100%  162.03MB/s    0:00:16 (xfer#6, to-check=0/8)

#1) sent 6.96G bytes  received 164 bytes  190.60M bytes/sec
#2) sent 6.96G bytes  received 164 bytes  244.10M bytes/sec
#3) sent 6.96G bytes  received 164 bytes  244.10M bytes/sec

=== LOCAL FS ===  rm -rfv /var/tmp/ISOS; rsync -a --verbose
--human-readable --progress ~mjoyner/ISOS /var/tmp/ISOS

#1) sent 6.96G bytes  received 164 bytes  167.63M bytes/sec
#2) sent 6.96G bytes  received 164 bytes  185.51M bytes/sec
#3) sent 6.96G bytes  received 164 bytes  220.85M bytes/sec

=== time umount /data
real    2m11.090s
user    0m0.000s
sys    1m49.566s


*********
=== *** mount -o sync,data=journal /dev/vgdrbd0/nfs1 /data *** ===

=== iozone test, no peer

real    0m9.483s real    0m9.583s real    0m9.583s real    0m9.559s
real    0m9.491s

=== izone test, w/ peer up.

real    0m41.558s real    0m41.386s real    0m41.456s real    0m41.448s


=== BIG FILES TEST
no peer) sent 6.96G bytes  received 164 bytes  74.40M bytes/sec
w/ peer) sent 6.96G bytes  received 164 bytes  29.79M bytes/sec
rsync/var/tmp2/var/tmp) sent 6.96G bytes  received 164 bytes  47.49M
bytes/sec

=== time umount /data

real    0m1.078s
user    0m0.000s
sys    0m1.075s

_______________________________________________
drbd-user mailing list
[email protected]
http://lists.linbit.com/mailman/listinfo/drbd-user



--

Michael Joyner ᏩᏯ
System Administrator/Edward Waters College
1658 Kings Road, Jacksonville, FL 32209
904-470-8170 (V) / 904-470-8170 (F)
ᏩᏙ

========================================================================
         HOW TO SUBMIT A WORK REQUEST TO INFORMATION TECHNOLOGY
------------------------------------------------------------------------

Via the web: http://otrs.ewc.edu/otrs/customer.pl

Via email: [email protected]

* It is imperative that you submit each problem as a new item.
* Please do NOT combine multiple problem reports in a single workorder.
* PLEASE DO NOT EMAIL SUPPORT STAFF REQUESTS DIRECTLY.
* PLEASE USE THE JOB TRACKING SYSTEM.

========================================================================


_______________________________________________
drbd-user mailing list
[email protected]
http://lists.linbit.com/mailman/listinfo/drbd-user

Reply via email to