alternative to SAN, please advise

[jman22]

Hello all!

I am looking for an alternative to SAN storage solution.

I am looking for a raid card that can stripe an array of 15 hard disks with raid 5 and then "share" the array with a raid card on another machine.

Let me explain, I have two application servers that will fail over in case of hardware failure and use load balancing. I would like the raid card from each machine be able to share a single array (I don't want to buy an additional 15 hard disks for the other app server). The problem I see with this is that when an array is built by one card with raid 5, parity bit conflicts will occur at some point and cause issues when the storage array is accessed/built by the other card.

So, I am looking for something like an expander back plane (like in a DAS) that could hold 15 hard disks, but has two, not one, Raid slots for sharing, and is intelligent enough to prevent the parity bit conflicts. Any one know of something like this?

Any help would be greatly appreciated!
Thanks!

 

[LeeMason]

Is there any particular reason why you don't want to use SAN storage? It is just expense or is there another reason?


Lee Mason
Optimal Projects Ltd

http://www.opl.co.uk

http://www.backupsanywhere.com

 

[jman22]

Lee, yes, expense. The quotes I'm getting for the HBA cards, switch, HDs, are more than I think they should be. Instead of adding a switch and raid controller cards on the SAN end for intelligence, I am trying to find a setup with intelligence built on the controller cards on the application server end (no switch unit, no additional raid controllers). I think what I want is out there...just have not looked in the right areas.

 

[xmsre]

fail over in case of hardware failure and use load balancing".


Which is it, failover cluster or load balancing? Two servers cannot have block level write access to the same drive at the same time. Failover clustering generally uses a device driver that implements the SCSI reserve command (clusdisk in Windows) to ensure one node or the other has exclusive access to the disk at any given point in time.

If you load balance, and each node is writing to the disk via block mode access methods, then you'll need two copies of the data and a mechinism to handle updates and conflicts. You could use a file access protocol like CIFS or NFS to access the data from both nodes, as long as the protocol has a mechinism to handle conflicts (oplocks in CIFS for example). Going further with the CIFS example, put the files on a CIFS file server and access them via a URL from each NLB member.

If you go the failover clustering route, then you probably want a to attach your a RAID array to a server with iSCSI target software installed. Connect to the server via the block mode iSCSI protocol and let the driver for your OS use SCSI reserve to control access to the disk.

 

[jman22]

xmsre, thanks for this information! My load balancing unit takes care of fault tolerance in case of hardware failure, that's what I meant. Looked at iscsi target software, a little pricey, and having two copies of 15 terabytes is not an option. I have played with designs using vmware, nas, das, and so forth, showed promise, but in all instances SAN is the better option.

I might have to go with SAN or split the data into 7 terabytes and replicate that way.

I'm doing a little more research into scsi and how it references memory blocks during read/writes, let ya'll know how it goes.

J

 

[xmsre]

Which OS are you running?

 

[jman22]

CentOS

Now I am looking into how much building my own SAN with redundancy would run me. Just realized I had this option.