[DRBD-user] DRBD performance (only uses 40% of a GE link)

Christian Balzer chibi at gol.com
Tue Mar 4 01:04:07 CET 2008

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,

On Mon, 3 Mar 2008 18:02:41 +0100 Lars Ellenberg wrote:
> On Mon, Mar 03, 2008 at 11:51:12PM +0900, Christian Balzer wrote:
> > > synchronous writes don't have the best latency characteristics
> > > with md raid5.
> > >
> > Yeah, but a RAID10 is just too... wasteful. ^^
> 
> you currently have 8 1TB disks in a md raid5 setup.
> you could have 4 smallish md raid1s in the first partition,
> and then still have the one big md raid5 on the rest of them.
> 
Oh, you meant for the metadata only... I already got the swap in a RAID10
for obvious reasons. Though while the writes to RAID10 would definitely be
a LOT faster, in the case of a massive write storm from the primary we are
still competing over the same 8 spindles. OTOH, we are doing that now
already, since the internal metadata is also not exactly adjacent to the
real data on those 2 3TB RAID5 (this is going to be an active/active
cluster of sorts, with each being primary for half of the storage by
default).
I guess I will give this a try, thanks for the input.

Regards, 

Christian.
-- 
Christian Balzer        Network/Systems Engineer                NOC
chibi at gol.com   	Global OnLine Japan/Fusion Network Services
http://www.gol.com/



More information about the drbd-user mailing list