[sheepdog-users] HW RAID or single disk

Matthew Law matt at webcontracts.co.uk
Thu Jul 19 11:07:05 CEST 2012

I think for most users there is a 'sweet spot' in configuration that best
matches their needs and budget.

If you use a HW RAID controller in a conventional, non-clustered
environment then most people would opt for a battery/super capacitor
backup to ensure writes are flushed in the event of power failure. Many
also choose to enable writeback cache too since the battery all but
guarantees data consistency and that gets them improved performance. 
Obviously HW RAID also has a number of disadvantages too, but I won't go
into those.

For clustered systems it often isn't so clear cut. Each user will likely
have different hardware, budgets and possibly data consistency minima.  In
the case of sheepdog, I don't think hardware RAID would provide much
advantage over, say, software RAID 1 since writes still have to be flushed
to the cluster to be classed as successful. It would certainly be an
expensive way of doing things as that RAID card might buy another node or
more sheep disks (more spindles implies better performance). And by using
any form of mirrored RAID level you are creating possibly many copies of
data/reducing the data density of your hardware.

You could look at it the other way too - if you have a small cluster (3
nodes or more) and only commodity hardware you might opt for copies = 2
and then use a RAID0 of two SATA drives for each sheep. Knowing that
sheepdog will not let you lose data by default.

Sheepdog is very flexible so you "pay's yer money and takes yer choice" :-)



More information about the sheepdog-users mailing list