[DRBD-user] 0.6 vs. 0.7 3ware performance? [Was: Expected syncspeed for a 425gb volume?]

Stephan Rattai srattai at zmnh.uni-hamburg.de
Wed Apr 13 09:08:26 CEST 2005

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


Hi there,

I also got two machines with 3ware controllers, one 8506-8 with 6 WD 1600JD 
drives and one with 6 Seagate ST3160827AS. I started with Raid 5 but got poor 
performance so I turned to Raid 10 which makes 

a) 70 Megs/sec while writing (block) and 95 in reading (also block) on the 
8506 and 

b) 71 Megs/sec while writing and 234 Megs/sec  on the 9500 (everything 
blockwise and measured through bonnie++ without drbd- on the scsi device 
itself).

I had a really sad performance at first (about 15-18 Megs per sec write 
performance with drbd) and took some steps to increase it. First there was 
the migration to Raid 10 which brought some performance. Then I moved the 
metadata to a separate disk which is not really funny as I do not know what 
happens when this one dies and drbd loses its metadata-device. This and the 
increase of /sys/block/sda/queue/nr_requests to 1024 as suggested by Philip 
Reisner some time ago on this list increased the performance to about 

a) 38 Megs/sec in writing and 58 in reading on the 8506 machine and
b) 30 Megs/sec in writing and 67 in reading on the 9500 machine on the drbd 
device. 

So the slower machine is faster in writing as the other end's lower device is 
faster there which really spots the bottleneck I guess. I don't know if you 
could call it decent performance, I guess next time I will go with the LSI 
800-8x. 

If you have the 9500 you should get the 9.2 software release, it turns on 
features like OCE and online Raid migration. The best part about it is a 
increase in performance (at least in my setup).

The sync from 8506 to 9500 makes about 50 Megs/sec, vice versa I get about 40. 

I also had a problem with one WD suddenly disappearing (timeout) - maybe I 
should try to get SD drives too. 

Am Mittwoch, 13. April 2005 02:36 schrieb Jeff Buck:
> On Tue, 2005-04-12 at 13:03 -0500, Nate Carlson wrote:
> > > back, write, and seek random again. The problem being that the 3ware
> > > raid card is splitting the writes into 64k chunks on the different
> > > disks, causing them all to do a little dance after every 64k write. We
> > > have a bunch of the "JD" drives.. I guess the ones that don't have this
> > > behavior are the "SD" drives (unavailable when we started this
> > > project).
> >
> > Makes a lot of sense - we've actually got two seagate and two WD (JD)
> > drives on each controller.
>
> I still suspect that there's a bit more going on with the 3wares though.
> I've yet to see anyone report any decent performance with them. Surely
> someone's got a setup without the peculiar WD drives. I'm thinking that
> there's still something about the 3wares or their drivers that doesn't
> get along with drbd at all. Maybe drbd 7 is more touchy about latency in
> the hardware?
>
> We've had problems with our WD "JD" drives dropping out of our array
> spontaneously also. One of the WD techs said that if we RA them, and
> tell them we're using them in a raid array they'll send us the "SD"
> version of the drive back. So far we've only got one of the drives back,
> but it's a nice shiny new "SD" drive this time, so it looks like he was
> correct. It might be an option for anyone that thinks this might be
> their problem.
>
> We've got 46 more drives to RA still, and they only let you do 5/day,
> one at a time, so beware.... It could take a while if you've got a lot
> of drives.
>
>
> _______________________________________________
> drbd-user mailing list
> drbd-user at lists.linbit.com
> http://lists.linbit.com/mailman/listinfo/drbd-user



More information about the drbd-user mailing list