[DRBD-user] Performance hit DRBD vs raw block device, even when disconnected

Lars Ellenberg lars.ellenberg at linbit.com
Mon Jun 4 11:29:20 CEST 2012

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


On Sat, Jun 02, 2012 at 02:02:03PM +0200, Wiebe Cazemier wrote:
> ----- Original Message -----
> > From: "Florian Haas" <florian at hastexo.com>
> > To: "Wiebe Cazemier" <wiebe at halfgaar.net>
> > Cc: drbd-user at lists.linbit.com
> > Sent: Friday, 1 June, 2012 9:03:42 PM
> > Subject: Re: [DRBD-user] Performance hit DRBD vs raw block device, even when disconnected
> > 
> > Sorry if I rant for a bit, but it's been _years_ since I wrote that
> > part of the User's Guide, and it's been regurgitated a million times
> > on this list, by myself and many others. Please, say it after me:
> > 
> > I will not use dd without oflag=direct or oflag=dsync.
> > I will not use dd without oflag=direct or oflag=dsync.
> > I will not use dd without oflag=direct or oflag=dsync.
> > 
> > You're kidding yourself. You're measuring *something*, a mix of
> > caching effects and actual I/O, but not the throughput of your block
> > device.
> > 
> > Please rerun your tests with oflag=direct, and then come back with
> > updated results.
> > 
> > Florian
> 
> Sorry, I should have realized that...
> 
> Testing now yields more normal results:
> 
> 1) about 25-30 MB/s on both nodes.
> 2) Higher speeds (about 50 MB/s) when DRBD device is in disconnect mode.
> 
> If I invalidate my test device (causing a resync), it syncs at about 60 MB/s, the speed I limited it to. I don't understand why normal throughput is so much lower. Fiddling with max-buffers, max-epoch-size and unplug-watermark doesn't really seem to influence the 25 MB/s much.
> 
> To give more information about the hosts:
> 
> Host1: Supermicro (mainboard: X8SIL-F). 3Ware 9650 SE SATA2 hardware RAID6, 5x 2TB disks. Xeon(R) CPU X3430, 2.40GHz. Machine is Xen dom0 and dom0 has 1 core.
> 
> Host2: Dell Poweredge R200. 2x250GB Linux MD RAID1. Xeon CPU X3220, 2.40GHz. Machine is Xen dom0 and dom0 has 4 cores.
> 
> Both nodes reach higher network and disk speeds then I'm getting here.

Put DRBD meta data on RAID 1.
Use decent battery backed (or otherwise non-volatile) write cache.
Disable barriers and flushes in DRBD.


-- 
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com



More information about the drbd-user mailing list