[DRBD-user] blocking I/O with drbd

Jan Škoda lefty at multihost.cz
Fri Dec 23 18:23:56 CET 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,

I have (almost) exactly the same problem as
http://lists.linbit.com/pipermail/drbd-user/2011-December/017405.html

Debian 6.0.2
DRBD 8.3.7, dual primary, protocol C
LVM

DRBD allows to write at high speeds (20MB/s first seconds), then the
speed falls and drbd blocks even after I kill dd. However when i run dd
with oflag=direct, write speed is about 10kb/s! The network latency is
quite high (0.200ms) and the systems are virtualized, so I haven't
enabled disk/md barriers/flushes, because I don't trust all the IO
layers under the DRBD.

Is there any way to shrink/disable that caching?

How should I tune the drbd network settings for better performance in
such a high-latency environment? (I mean especially those buf sizes.) As
you can see from the pastebin entries (below), load is quite high even
when idle.

How does file access with DRBD & OCFS2 work?  I suppose that a node can
read a file without having to interact with the second one (without
locking).

Example of idle behavior:
http://pastebin.com/raw.php?i=YjQW05Us
During dd writing with oflag=direct on another node:
http://pastebin.com/raw.php?i=gvCQkBQ7
You can get the dstat plugin on my web:
http://lefty.ofight.org/howto/drbd-dstat-plugin



More information about the drbd-user mailing list