Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hallo together,
I'm just struggling with the same issue.
With drbd 8.0.6, CentOS 5.1,
on IBM xSeries x3650, ServeRaid-8k-l (Battery), Raid-6, 6x300GB/15k-Disks and Gig-Ethernet
I get ~90 Mbytes/sec write-throughput with /dev/drbd stacked on a LV.
>>> optimal, only limited by ethernet.
(Without DRBD: 200 - 300 Mbytes/sec)
With drbd 8.2.6 im stuck to 18.0 Mbytes/second
or with new parameters no-disk-flushes and no-md-flushes
ist still limited to: 74 MB/s
So there's still some loss of speed between 8.0.8 and later versions?
Be careful: for small data quantities below 4 Gbytes the degradation
is not as visible!!
Anyone any idea?
Tested with:
-------------
drbdadm primary test;
cat /proc/drbd;
for i in rootvg/4direct drbd1 ; do echo /dev/$i:; dd if=/dev/zero of=/dev/$i bs=1M count=4096; done
/dev/rootvg/4direct:
4096+0 records in
4096+0 records out
4294967296 bytes (4.3 GB) copied, 19.4024 seconds, 221 MB/s
/dev/drbd1:
4096+0 records in
4096+0 records out
4294967296 bytes (4.3 GB) copied, 238.107 seconds, 18.0 MB/s
disk {
on-io-error detach;
no-disk-flushes;
no-md-flushes;
}
===>
/dev/rootvg/4direct:
4096+0 records in
4096+0 records out
4294967296 bytes (4.3 GB) copied, 19.5529 seconds, 220 MB/s
/dev/drbd1:
4096+0 records in
4096+0 records out
4294967296 bytes (4.3 GB) copied, 57.6269 seconds, 74.5 MB/s
Jörg
-----Ursprüngliche Nachricht-----
Von: drbd-user-bounces at lists.linbit.com [mailto:drbd-user-bounces at lists.linbit.com] Im Auftrag von Philipp Reisner
Gesendet: Samstag, 24. Mai 2008 21:35
An: drbd-user at lists.linbit.com
Betreff: Re: [DRBD-user] 8.x performance vs 0.7x --- was (not urgent) Requestfor DRBD Developers
[...]
> >> Personally, I'm still on 8.0.8. Later versions at the time seemed
> >> to show _VASTLY_ reduced performance on my hardware, so I excluded
> >> it from the updates. I'm not experiencing any problems, so I'm sticking with it.
That is due to that change:
8.0.9 (api:86/proto:86)
--------
* In case our backing devices support write barriers and cache
flushes, we use these means to ensure data integrity in
the presence of volatile disk write caches and power outages.
[...]
8.0.12 (api:86/proto:86)
--------
[...]
* Two new config options no-disk-flushes and no-md-flushes to disable
the use of io subsystem flushes and barrier BIOs.
[...]
But please *read* the documentation about "no-disk-flushes" and understand the implications before you set these options.
# In case you are sure that your storage subsystem has battery
# backed up RAM and you know from measurements that it really honors
# flush instructions by flushing data out from its non volatile
# write cache to disk, you have double security. You might then
# reduce this to single security by disabling disk flushes with
# this option. It might improve performance in this case.
# ONLY USE THIS OPTION IF YOU KNOW WHAT YOU ARE DOING.
# no-disk-flushes;
# no-md-flushes;
-Phil
_______________________________________________
drbd-user mailing list
drbd-user at lists.linbit.com
http://lists.linbit.com/mailman/listinfo/drbd-user