Sharing Disks Between Two Servers?

[SeanPattersonTC]

I was looking to basically cluster two DL360 G5 servers for the purposes of storage. Here is the scenario: 

Server A has 6 disk drives. 

Server B has 6 disk drives. 

I would allocate two drives from Server A for its own OS. 

I would allocate two drives from Server B for its own OS. 

Then I want to take the remaining four drives from each server and make ONE logical drive out of all eight of them so that BOTH servers could share it as one drive. 

Can this be done if I put fibre channel cards in both servers and join them together? Or some other connection between the two? 

THANKS!

[Torsten.]

You need some external drives, e.g. a P2000 array or a MSA2040 in order to share disks between 2 servers. You cannot share server internal drives.

[anthony11]

Sure you can.

GlusterFS

HDFS

Ceph

PVFS

One could even do various perverse things with NFS and transluscent filesystems but that wouldn't exactly be what was requested.

[Torsten.]

You can do a lot of things, but the basic idea of a cluster is to keep it working if a node is down.

Unless you re-invent a VMS-style cluster, you should use external disks.

[Matti_Kurkela]

> Then I want to take the remaining four drives from each server and make ONE logical drive out of all eight of them so that BOTH servers could share it as one drive.

Which OS would you use?

Should the data be mirrored between servers or otherwise made redundant, so that if one server is down the other could keep working? Or are you willing to accept the logical drive being inoperable unless both servers are alive and well???

What is the actual problem this set-up is supposed to solve?

On Linux, you won't even necessarily need FibreChannel HBAs: you can achieve the same thing with regular network cards and DRBD.

http://www.drbd.org/

Note that DRBD only provides the disk device layer. To successfully use the disks as a filesystem on multiple hosts simultaneously, you will need to choose a filesystem type that allows such use.

The usual filesystems have a designed-in assumption that there are no other hosts accessing the filesystem at the same time. This allows efficient caching. If this assumption is violated, the per-host disk caches will eventually cause the applications to receive stale data, even if only one host has read/write access at a time.

(The read-only host caches the data it has read previously, and won't be informed that the changes made by the read/write host has invalidated the data in the cache of the read-only host. When this happens to filesystem metadata, the read-only host is likely to read the wrong blocks for at least some subsequent requests, resulting in applications receiving nonsensical or subtly corrupted data.)

For multi-host access, you will need a cluster filesystem, which coordinates the write operations across all the participating hosts.

[SeanPattersonTC]

It is all Windows Server 2008 R2. 

Here is what is transpiring. I have an application installed and running on Server A. I want to add Server B and configure as a cluster. In order to cluster, the application installed on both server would need to share some common directories. I've got a LARGE amount of storage space already in Server A and identical in Server B so I was trying to avoid having to purchase a SAN since I've already acquired plenty of storage space. 

I would need the logical drive to stay operable in the event that the entire other server goes down. Would likely do a RAID 0 +1 scenario and split the mirrors across each server. 

However, it sounds like for the aggrevation, I might just try to pick up a SAN and go that route.