[DRBD-user] no-disk-flushes ineffective?

[Gordan Bobic]

On Wed, 18 Mar 2009 21:57:38 +1000, "Andrew (Anything)"
<anything at starstrike.net> wrote:
> Hi Gordan.
> 
> Im pretty sure I tried no-disk-drain also in desperation. Ill redo it
with
> some benchmark results soon for you.
> Its an almost unmodified default copy. Ive cut out all the comments for
> here.
> 
> 
> common {
>   syncer {
>         rate 100M;
>   }
> }

Try rate 10M here if you're on a 100Mb network. Rate is in MB not Mb.

>   disk {
>     on-io-error   detach;
> 
>     no-disk-barrier;
>     no-disk-flushes;
>     no-md-flushes;
> 
>   }

Add no-disk-drain here and see if it makes a difference.

>   net {
>     sndbuf-size 0;
>     allow-two-primaries;
>     cram-hmac-alg "sha1";
>     shared-secret "xxxxxxxxxxx";
>     after-sb-0pri discard-younger-primary;
>     after-sb-1pri discard-secondary;
>     after-sb-2pri disconnect;
>     rr-conflict disconnect;
>   }
>   syncer {
>     rate 100M;
>     al-extents 1201;
>   }

See above re: rate.

Have you checked what sort of transfer rates you get over the network
interface (e.g. NFS, ftp, netcat on small and large files)? What
switch/NICs are you using? What speed do you get out of it when the devices
are re-syncing (cat /proc/drbd during resync)?

Gordan