[DRBD-user] DRBD performance and heavy load

Lars Ellenberg lars.ellenberg at linbit.com
Sun Mar 9 23:29:09 CET 2008

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


On Sun, Mar 09, 2008 at 09:58:14PM +0100, Mario Peschel wrote:
> Hi Sam,
>
> I tried the newest kernel 2.6.24 with the kernel module mptsas, without  
> any success.
>
> Then I found the following solution:
>
> http://lists.us.dell.com/pipermail/linux-precision/2007-June/001188.html

which boils down to "enable write cache" on the disks.

I don't know about your controller, but I do hope
that cache is backed by a good battery.  really.
otherwise you risk your data in face of power outage.  

> Now my performance is much better:
>
> # dd if=/dev/zero bs=4096 count=10000 of=/tmp/testfile oflag=dsync

WCE=on
from main memory to disk cache.
basically you measure the bus latency here.
> 40960000 bytes (41 MB) copied, 7.02387 seconds, 5.8 MB/s
~latency per request: 0.7 ms


WCE=off
this is the latency of getting 4kB
from main memory all the way down to rotating rust.
>>>  40960000 bytes (41 MB) copied, 263.415 seconds, 155 kB/s
~latency per request: 26 ms

without cache, mostly depending on rotation per minute,
disks should theoretically have some latency of 3 to 10 ms.
so unless you have very slow disks, there is aparently some logic with
timers involved, arbitrarily delaying single requests, "anticipating" more.
all tuned for maximum bandwidth, not minimum latency.

-- 
: commercial DRBD/HA support and consulting: sales at linbit.com :
: Lars Ellenberg                            Tel +43-1-8178292-0  :
: LINBIT Information Technologies GmbH      Fax +43-1-8178292-82 :
: Vivenotgasse 48, A-1120 Vienna/Europe    http://www.linbit.com :
__
please use the "List-Reply" function of your email client.



More information about the drbd-user mailing list