On Saturday 11 December 2010 18:37 John R Pierce wrote
On 12/11/10 9:29 AM, Rafa Grimán wrote:
What about a DIY NAS with an off the shelf server and storage array?
and how do you avoid single-point-of-failure? if that COTS goes down, your storage is offline, and you've lost any writes in progress.
enterprise storage has fully redundant *everything*, a dual filer HA NetApp box will not even blink if any subsystem fails. cached write data is mirrored in both controllers so if one controller croaks, the other can flush the buffers for it, and client storage connections continue without a hiccup. yes, you pay for this.
As I said a bit further down in my previous e-mail: you can use GPFS, CXFS, GFS as its filesystem and have redundant hardware.
If you want to get a better idea, check out SONAS from IBM. It's basically GPFS + Samba and redundant HW. You get HA, high performance, modularity and flexibility. You can also check out:
http://www.redhat.com/magazine/006apr05/features/gfs/
If you don't want to shell out what IBM is going to charge you, you can DIY (if you have the knowledge) or you can hire some Linux consultant that has a good Linux knowledge.
HTH
Rafa