[Linux-cluster] Redundant (multipath)/Replicating NFS

isplist at logicore.net isplist at logicore.net
Sat Feb 16 00:10:04 UTC 2008


> If your NFS servers are proprietary hardware, I doubt you'll find
> anything to handle replication between the two of them, unless the
> appliance vendor has a module available for such a thing.

They can back up to each other's hardware, things like that but I was thinking 
more along the lines of something which could happen from the web servers 
themselves. Perhaps writing to two locations for every write or something.
 
> So you have 2 FC chasis with 12 drives each? I doubt mirroring them is
> vital provided they are running RAID5 or RAID6 on each.

Each controller chassis is the RAID chassis. At this point, I've been using 
RAID5 (5,3) since I don't need the extra speed of (0,0+1) for writes since 
it's for the mainly static web pages. So the RAID5 is spread across the 12 
drives on each chassis. Then on each controller, I have two additional chassis 
to extend the drive loop on each controller. So basically, the controller is 
taking care of 3 sets of RAID5 drives. Each chassis has dual loop controllers 
and the RAID chassis has dual loop and RAID controllers. Of course, each 
chassis has dual power supplies and it's all on UPS/Generator. 

I then connect the RAID controllers from each RAID chassis into an FC hub. The 
NAS servers see all of the storage and I just make it available as NFS through 
the NAS server. Ends up being a lot simpler than what I had going before. The 
cluster was overkill for the web pages.

> fails, you'll still have downtime until you can replace it, even if you
> can transplant all the drives into a new chasis. It depends on how much
> downtime you can live with, and on whether you need a system with
> absolutely no single points of failure.

Well, and this is where I'm not sure about industry standards. I know we all 
want zero down time but when cost is an issue initially, how do most folks go 
about it? Can LAMP services where users store media have an acceptable down 
time limit until they reach a certain point? The hardware I've chosen seems to 
offer a lot of security but as you point out, there could definitely be some 
down time if I have to move drives around.
 
> I don't think near-line storage is going to be workable for on-line
> media serving.

I was thinking more in terms of safekeeping the data, not serving it up from 
there. But then, if things get busy, that still would not be a great solution 
since tape is so slow and the constant backups would be a headache.

> headroom you need. Based on this you might decide to run RAID 01 or RAID
> 51 on the setup. But if you need mirroring for the NFS proxies, I
> suspect that's only going to be doable if the appliances support it.

More than anything, I am trying to find acceptable growth paths that will 
allow for budget conscious methods of getting to the point of being able to 
afford all the redundancy in the world. I can't imagine what youtube does?

Mike






More information about the Linux-cluster mailing list