[bjames]

Can an external scsi raid be shared between 2 systems?  

I got this idea of sharing a 500gig Medea raid connected to 2 sgi computers from my E-mu sampler set up where I have a external scsi hard drive and cd-rom between 2 E-mu keyboard samplers.  This set up is supported as long as you don’t have both samplers trying to write to the disc at the same time.

[weblacky]

IF, you can actually perform a valid drive/enclosure hookup for that (like using Fibre FC-AL or the have an enclosure that support multiple drive paths).

Yes, you can…provided you use either special file systems or tools that prevent simultaneous access!!!

For example, if you wanted two computers (identical models and specs) to share a booting OS (only one at a time, like a hot spare). You can offline the drive on one machine and online it on another and go! That works for some failover scenarios. Just have a system ready to turn on if the other system dies for a power supply or overheating or cooling reason, you can just turn off the failing system and start up the spare system and it’s already hooked to the same drive that the other one was prior.

Now, if you want simultaneous system access to the same drive, you have to use a clustering file system. There was one for SGI, CFXS, it was a separately license product.

The clustering file system is what orchestrates host access to the file system of a volume using either a network link or the file system itself. I have seen implementation of both types of clustering file systems.


Normally, this is done for a multi target, raid enclosure, or a Fibre SAN.

For a single disc this would likely just go very slow and and just put wear on the drive.


But to answer your question simply, it is absolutely doable if the hardware supports it. You must use a special cluster file system that still orchestrates rewrite access, so it is sequential and not simultaneous to the underlying hardware. Irix did have one, but I thought it was a separately paid product that I haven’t heard of anyone hacking.

https://www.pircs.iastate.edu/WhatsNew/i....html#cxfs

[bjames]

Hmmm.. my set up would be as follows:

Octane —> Medea raid —> O2. Assuming the computers have an integral termination (like the E-mu samplers).

The Medea raid has 2 scsi ports, input and output for either connecting more HDs or terminating.

[weblacky]

That doesn’t work. You cannot connect both ends of a parallel SCSI chain to a host. You’d have to have an enclosure that emulates multiple targets. Other forms of SCSI have multiple paths (fiber 40-pin SCSI does). But yeah, natively parallel SCSI is single host for a chain (bus).

[robespierre]

What weblacky said is, as far as I understand it, all correct, but there is a lot more to this topic.
If you just connect a SCSI drive to two host computers with SCSI cables, the shared bus scenario, then you have some additional problems. One of them is that the bus will get reset each time either computer reinitializes it. So you can't actually be using the disk from Host 1 at the same time you reboot Host 2. The other issue is that the two hosts need different addresses on the bus. And the same issue as above that with XFS, you can't share access to a volume, so the two hosts can't mount the same volume at the same time.
SCSI-2 (and above) define how the target must keep requests from different initiator IDs separate and how multiple initiators arbitrate for bus access.

[weblacky]

I was “into” SANs and SCSI target emulation for a few years…youth.

If you’re looking for a cheap way of getting multipath SCSI, you need to try something like SCST For Linux.

With the right hardware you can run SCSI target emulation to the same “virtual volume” from a several old Parallel SCSI controllers. Often requires a custom kernel!

But allows a magic box with any number of disks to become a virtual SAN for either Fibre FC-AL and/or SCSI SE targets.

It’s more work than cost. This suggestion DOES NOT lift the requirement for a clustering (shared) file system in any simultaneous host to volume access scenario! Still very much a base requirement.

So for Irix, specifically, unless you “come up with” a CXFS license for all the SGIs involved, all the hardware in the world won’t make it work.

FYI, Linux should have a free clustering file system solution for itself, but not for any other OS, that I’m aware of.

I think you should consider a shared network folder (via NFS) and likely go from that. Physical disk sharing isn’t well supported freely, often a solution is special purpose and it’s not that much better/faster than a gigabit network to fast shared volume.

Also Irix does not support iSCSI…so that won’t work either.