is there a distro that allows clustering and basic san ?

[aberant]

Hi there. So i've found cluster knoppix and that seems brilliant. There's only one other thing that would be nice, and that's to have the cluster work as a s.a.n. so that the hd space could be used as a share on a network. Hope that makes sense..... any ideas ?

 

[drag]

I think I know what your talking about...

Like take a bunch of computers and a bunch of harddrives and combine them into one big filesystem (with replication and duplications and locks and stuff like that to ensure data integraty and failsafe operation and all that)?

For example you have the coda distributed file system, that could be one...

OpenMosix has oMFS for distributing data between the different proccesses... There are others, MS has one, sortoff. That's when you use the UNC's to navigate around.

I don't know that much about this sort of stuff, I still have issues configuring SAMBA time to time.

here is some definitions, maybe this can help you explain what you are looking for.

 

[aberant]

i shall rephrase. a distro that i can install on a bunch of machines to make it appear to a user (my gf for example) that there is one drive sized 10 tb or whatever when infact its 20 machines each with 4 250gb hard drives.

 

[Sianath]

You can configure a Server 2003 machine as an iSCSI target if you like, but if you are looking to use servers as consolidated storage in a SAN environment, you are going about it the opposite way the rest of the industry is.

Machines (no matter what they are running) are much more prone to error than any SAN target is, and on top of that you would only have a JBOD setup at best (assuming you decided to farm a bunch of Server 2003 machines as iSCSI targets). Also, each of your iSCSI targets would be a separate target. It takes SAN software + hardware (read: SCSI controllers) to consolidate separate data stores into a single logical disk that is transparent to the OS.

The closest you could get in Windows (if you wanted to emulate a SAN environement) would be a bunch of 2003 servers as iSCSI targets and have all those targets set up as dynamic disks with some form of software RAID set.

 

[drag]

i shall rephrase. a distro that i can install on a bunch of machines to make it appear to a user (my gf for example) that there is one drive sized 10 tb or whatever when infact its 20 machines each with 4 250gb hard drives.

There is no distro that sets that up as a default, but it's possible.

Linux has the LVM support which can do the single logical volume stuff thru scsi channels and whatnot, but not across network stuff.

here is a oracle paper on CVS and SLVM

CVS is the Veratis Cluster Volume manager, and SLVM is HP's Shared Logical Volume Manager.

But I don't know if those have linux support or not, or are they only for stuff like HP-UX Unix? Either way they are closed source and very expensive.


OpenAFS is from IBM and has linux support
Coda is a experimental high performance distributed file system
Intermezzo is another experimental file system inspired by Coda

These things work by having a "unified namespace" and is divided up into volumes.

So that you just mount /coda (for instance) and then entire network directory is there as one gigantic filesystem.

For replication and performance you can have many copies of the same volume on bunches of different servers, so that if one was to go down then the clients will still be able to access their files like nothing happened.

However each volume is smaller then the actual partitions your harddrives use. That way they can be moved around from partition to partition easily.

So if you had all the volumes as 100gigs you couldn't put a 200gig file in there, but if you had 20 machines serving 250gig drives you could still have 5000gigs worth of information floating around. To any client it would seem that they would be accessing one 5000gig (probably much less because of all the overhead for metadata and stuff) file system, just divided up between different volumes. Or if you had 20 machines running 250gig drives, then you could replicate the volumes between all the servers and have 2500gig worth of information, but have it like a gigantic RAID 0 array..

So that's sorta like what you want.... It would appear to the end user that you would have one gigantic server your running off of, but all the files would be distributed between all the different smallish volumes (themselves appering like directories in a file system) which would be distributed across all the individual file servers.

It's sort of like the traditional Unix file system, were you have mount points which are directories. You could stick / and all the file systems on one big partition, or you could put /boot on one partition, then /usr on another, and maybe mount /home on a remote NFS server. To the end user it would appear to be the same.

So stuff like Coda would be a extension of that. Traditional distributed file servers like NFS or SMB (windows file and print sharing) are like Windows C:, A:, D: drives...

examples of AFS being used successfully