Mailing List Archive: drbd replication speed

Hi,

I can't seem to get drbd to use any speed other than ~500Kb/sec for
resynchronisation after issuing a drbdsetup /dev/nb0 repl command.

in the config file I have

OPTIONS="--sync-rate 5000 -d 10241248 -p"

also tried it which "-r" isnstead of --sync-rate, no change. The option
doesn't seem to have any effect whatsoever, 1000 instead of 5000 doesn't
change a thing.

The network is switched 100Mbit, benchmarking gives a write performance of
about 8MB/sec. Limit is Harddisk performance, the hardware raid5 cna't do
more that that.

Any idea what's wrong?

Bye, Martin

PS: using 0.5.6 from CVS dated may 20 2000.

"you have moved your mouse, please reboot to make this change take effect"
--------------------------------------------------
Martin Bene vox: +43-316-813824
simon media fax: +43-316-813824-6
Andreas-Hofer-Platz 9 e-mail: mb@example.com
8010 Graz, Austria
--------------------------------------------------
finger mb@example.com for PGP public key

Am Fre, 26 Mai 2000 schriebst Du:
>Hi,
>
>I can't seem to get drbd to use any speed other than ~500Kb/sec for
>resynchronisation after issuing a drbdsetup /dev/nb0 repl command.
>
>in the config file I have
>
>OPTIONS="--sync-rate 5000 -d 10241248 -p"
>
>also tried it which "-r" isnstead of --sync-rate, no change. The option
>doesn't seem to have any effect whatsoever, 1000 instead of 5000 doesn't
>change a thing.
>
>The network is switched 100Mbit, benchmarking gives a write performance of
>about 8MB/sec. Limit is Harddisk performance, the hardware raid5 cna't do
>more that that.
>
>Any idea what's wrong?

Nothing wrong Martin,
it's (current) DRBD. On Intel (HZ=100) the theoretic maximum sync rate is
3200KB pr second. (On Alpha HZ=1024 it's 32MB/s..)

(HZ is a constant in the kernel sources, that says how often the
timer interrupt is raised in a second)

You are not reaching the 3.2MB/sec because the syncer is only using
a single page as buffer, and your hard disk can not deliver the data
instantly.

There are two ways to improve the situation:
*) Using more buffer-space (32-64 KB)
*) Using a bigger socket buffer.

This is going on the DRBD-TODO list.

-Philipp