[Lustre-discuss] DRDB transfer rate (was: Lustre as a reliable f/s for an archive data centre)

[Heiko Schröter]

On Donnerstag, 9. April 2009 08:47:00 Adam Gandelman wrote:
> Kevin Fox wrote:
> > If you striped 1, and gathered the metadata like I do with the e2scan
> > patch, you could checksum the data directly on the oss's. I've been
> > meaning on writing a system like this at some point (And actually have
> > had to do it manually once, in a disaster) but haven't had time to yet.
> >
> > As far as raid1 pairing nodes, you might be able to hack something
> > together using drbd and oss failover. No clue if its been tried before.
>
> I've just finished putting together a basic Lustre cluster in a lab
> using DRBD for redundancy and heartbeat for failover on both MDS and
> OSS.  Failover is working fine on both.  We're using mostly commodity
> hardware.  So far the setup seems like a solid candidate for a very
> cost-efficient way of bringing redundancy and high-availability to
> Lustre.  In the coming weeks  we'll be growing the cluster as we do
> benchmarking and hopefully have something to add to the DRBD section of
> the Lustre  wiki(which is pretty thin ATM)
>
Did you have the chance to measure the data transfer inside DRBD ?

We did observe it can be pretty slow (5MB/s). But might be related to a 
misconfiguration and or hardware issue though ...

Heiko