Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hi,
I'm trying to improve the performance of a DRBD (8.X) cluster machines I
did just set up.
Performance over the DRBD partition:
$ dd if=/dev/zero of=./tmp bs=4096 count=1000000
1000000+0 enregistrements lus
1000000+0 enregistrements écrits
4096000000 octets (4,1 GB) copiés, 84,7843 seconde, 48,3 MB/s
Performane over the local raid array:
$ dd if=/dev/zero of=./tmp bs=4096 count=1000000
1000000+0 enregistrements lus
1000000+0 enregistrements écrits
4096000000 octets (4,1 GB) copiés, 42,9673 seconde, 95,3 MB/s
Nearly twice as much on the same array/controller without drbd than with
drbd.
Here are the sysctl entries:
net.core.rmem_max = 16777216
net.core.wmem_max = 16777216
net.ipv4.tcp_rmem = 8192 174760 2097152
net.ipv4.tcp_wmem = 8192 174760 2097152
net section of drbd.conf:
sndbuf-size 240k;
max-buffers 20480;
unplug-watermark 20480;
max-epoch-size 16384;
syncer section of drbd.conf:
rate 90M;
al-extents 257;
Do you guys have any idea of the bottleneck of that setup ?
Thanks
Laurent