[DRBD-user] metadisk internal/external performance

Patrick Westphal patrick.westphal at inqbus.de
Thu Jun 23 10:56:59 CEST 2011

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


Hello DRBD users,


to measure the gain of an external metadisk I have done some performance
tests with two different setups. Setup A having internal metadata an a
RAID 10 (SAS) and Setup B with additional RAID 1 of SSDs used as
metadata disk:

  Setup A:
  Data : SAS RAID 10
  Metadata : Internal SAS RAID 10

  Setup B:
  Data : SAS RAID 10
  Metadata : External SSD RAID 1


In both setups I used the same LSI controller with 512 MB RAM.

For the tests I restricted the available kernel memory to 384 MB via mem
boot param to avoid measuring the linux cache performance. :)

I did not tweak the standard drbd.conf that much, except that I set a
higher syncer rate.

I can see an obvious difference if I run the test on the standalone
devices (SAS RAID 10, SSD RAID 1), e.g.

  SAS RAID 10 without DRBD (rate and latency):
  root at helios:/mnt/sas# tiotest -t 8 -f 2000 -r 20000
  [...]
  | Random Write 625 MBs | 115.6 s | 5.404 MB/s | 0.7 % | 3.7 % |
  [...]
  | Random Write | 5.335 ms | 4713.499 ms | 0.06813 | 0.00000 |
  [...]

vs.

  SSD RAID 1 without DRBD (rate and latency):
  root at helios:/mnt/ssd# tiotest -t 8 -f 2000 -r 20000
  [...]
  | Random Write 625 MBs | 12.6 s | 49.643 MB/s | 0.8 % | 18.0 % |
  [...]
  | Random Write | 0.591 ms | 4022.635 ms | 0.00500 | 0.00000 |
  [...]

  -> The random write rate is ten times higher as well as the latency is
ten times lower for the SSDs.

there is nearly no difference between the internal/external metadata
setups. Using the same tests as above I only get a performance
improvement of 10% (concerning random write):

  Setup A (metadisk internal, rate and latency):
  root at helios:/mnt/drbd# tiotest -t 8 -f 2000 -r 20000
  [...]
  | Random Write 625 MBs | 112.8 s | 5.543 MB/s | 8.7 % | 193.6 % |
  [...]
  | Random Write | 5.063 ms | 5566.188 ms | 0.02562 |  0.00000 |
  [...]

vs.

  Setup B (metadisk external, rate and latency):
  root at helios:/mnt/drbd# tiotest -t 8 -f 2000 -r 20000
  [...]
  | Random Write 625 MBs | 102.3 s | 6.110 MB/s | 2.3 % | 109.5 % |
  [...]
  | Random Write | 4.558 ms | 4089.111 ms | 0.04813 |  0.00000 |
  [...]

I already read Florian Haas' blog post
(http://fghaas.wordpress.com/2009/08/20/internal-metadata-and-why-we-recommend-it/)
and also got the points made in february on this list, but I'm wondering
if the effect of (not) using external metadata is really that small or
if I made some serious mistakes as far as my test setup is concerned.


Tanks in advance

Patrick
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20110623/f842bd48/attachment.htm>


More information about the drbd-user mailing list