DRBD

[Flaherty, Patrick]

> On June 24, 2008, Mark Komarinski sent me the following:
> > Anyway, DRBD is self-contained and is automatic (if the 
> remote system
> > disappears, it'll automatically resync when it reappears). You can
> > encrypt the stream, you can enforce how fast it syncs, and setup is
> > pretty easy.
> 
> Is this a Linux only project, or is it portable to other operating
> systems?
> 
DRBD is linux only (sorry FreeBSD). 

It's curious to me when I hear folks say it's been surpased by gfs. They
are different beasts. DRBD is for high availablitiy setups, GFS for
clusters. DRBD is a block device, GFS is an actual filesystem. High
availablity is when a service can't go down, clusters are used for
highly parallal tasks. It's nice when the service that can't go down is
highly parallal and fits in a clustered steup, but that is rather rare. 

You can run mysql on gfs if you use external locking, but I've read
terrible things about performance, and having more than one database
engine operating on the same data gives me the willies. For our app, low
writes, high reads, 2 node drbd works great. If we needed to scale out,
we'd use readonly replciated slaves from the HA head node.

Patrick