Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hello Lars and all,
> well...
> if it does, its a serious bug.
> it did not happen to me so far.
switching the protocols worked fine :)
By switching to protocol B we did not gain a noticeable performance increase,
but after switching to protol A
"iozone -s 1g -r 1m -o -i 0" (not via nfs but directly on the server) now
shows 16-18MB/s, instead of the previous 10-12MB/s. More important for us are
the nfs-client numbers (measured with 'time cp /local_disk/... /home/...',
which will happen in practice very often on our systems):
1 client: 9 - 10 MB/s
2 clients: 7.5 - 8 MB/s
4 clients: 4 - 4.5 MB/s
(/home exported with the sync option)
Thats still not what we got with 2.6.7 (4x7.5MB/s) but still much better than
what we get with protocol C and 2.4.27-rcX.
With protocol C we also had very high latency numbers, which e.g. slowed down
compiling of projects dramatically. With protocol A this problem is also gone
away :)
>
> what you can do to try and tune drbd:
> play with sndbuf-size
> increasing it may help throughput,
> decreasing it may help latency,
> ... or vice versa ... it depends ...
> play with the mtu of the link (jumbo frames)
> play with max-buffers
Well, I played with all of it (max-buffers, max-epoch-size, sndbuf-size) but
it didn't help/change so much.
The current values are:
sndbuf-size 524288;
max-buffers 16384;
Cheers,
Bernd
PS: Lars, do you want me to redo the latency tests, now with protocol A?