Contributor
Contributor

Cluster in a box - simple

Jump to solution

I am attempting to setup a cluster in a box.

I have a server running ESX 3.0.2 update 1, and 2 VM's that are server 2003 Enterprise SP2. I have setup the C drives for both VM's on local storage, SCSI adapter 0:1 on.

I have also setup 2 extra VMDK files on local storage, and zeroed them out. I connected those using a new SCSI adapter 1:1 and 1:2 on each VM, and set the bus to be shared for that SCSI adapter.

I can see the drives from within each Windows VM, and I want to test before I setup clustering. Just for a test, I created a file on one of the shared drives from one node, and looked to the other node expecting to see that same file showing up, and it does not. Is this normal, or do I have my setup wrong? This is my first stab at this of course.....

Any help or comments would be appreciated!

0 Kudos
1 Solution

Accepted Solutions
Virtuoso
Virtuoso

NTFS is NOT a clustered file system. You cannot expect to connect two machines to the same Disk/LUN and read and write from both at the same time.

Microsoft Clustering Services acts as an arbitrator between two hosts to ensure that there will only ever be one host that has access to a drive at any one time. It uses SCSI reservations to achieve this. If you have not yet set up the drive as a cluster resource, you will see the behaviour you're getting. This is not virtualisation-specific behaviour.

Warning - if you give more than one host access to a drive in the way you have (that is, without having MSCS act as arbitrator), you will get data corruption eventually. There are no checks and balances to ensure they do not tread on each other's toes. (Host A reads directory structure into cache, Host B modifies directory structure, Host A does the same thing but flushes modified copy from cache - you get the idea.)

View solution in original post

0 Kudos
8 Replies
Immortal
Immortal

Here's some good cluster guides:

Setup for Microsoft Cluster Service - http://www.vmware.com/pdf/vi3_vm_and_mscs.pdf

Guide to creating a vmotion-able Microsoft cluster with VI3 - http://www.vmware.com/community/thread.jspa?messageID=625404

Some musings on MSCS limitations - http://www.vmware.com/community/thread.jspa?threadID=83511

-=-=-=-=-=-=-=-=-=-=-==-=-=-=-=-=-=-=-=-=-=-=-

Visit my website:

-=-=-=-=-=-=-=-=-=-=-==-=-=-=-=-=-=-=-=-=-=-=-

0 Kudos
Expert
Expert

I think you will find that is normal....the shared disk, as a cluster resource, is owned by one node at a time.

Chris

0 Kudos
Contributor
Contributor

I did read through this one "Setup for Microsoft Cluster Service - "

Basically I followed all of the steps in there. Right now, both VM's can see all of the disks, even the shared ones. At the same time. When I make a test file on the shared E drive on node A, and go to node B and look at the shared E drive, I expect to see the test file there, but it's not....is that normal? I just don't want to go through the work of making the cluster if that is not normal behavior.

0 Kudos
Hot Shot
Hot Shot

That behavior is what we saw when we tried it, so I guess you could say it is normal. We were trying to create a Single Copy Cluster for Exchange 2007. After running into several isues (including the one you describe) we gave up and switched to hardware. It may be possible to do it, but we couldn't find any clear documentation on how. It could be that you need to set up the cluster in windows before the shared drive will work properly. Maybe someone else here in the forums can give us insight into how to make it happen.

0 Kudos
Enthusiast
Enthusiast

Yeb, this is a MS Issue.

Safe file on node 1, check node 2

then you have 3 options

1) You can see the file -> OK

2) You cann't see the file -> solution -> rescan disks again solves issue

3) You cann't see the file -> solution -> just close and open de diskmanager again on node 2 also solves issue

Cheers,

Jan

0 Kudos
Virtuoso
Virtuoso

NTFS is NOT a clustered file system. You cannot expect to connect two machines to the same Disk/LUN and read and write from both at the same time.

Microsoft Clustering Services acts as an arbitrator between two hosts to ensure that there will only ever be one host that has access to a drive at any one time. It uses SCSI reservations to achieve this. If you have not yet set up the drive as a cluster resource, you will see the behaviour you're getting. This is not virtualisation-specific behaviour.

Warning - if you give more than one host access to a drive in the way you have (that is, without having MSCS act as arbitrator), you will get data corruption eventually. There are no checks and balances to ensure they do not tread on each other's toes. (Host A reads directory structure into cache, Host B modifies directory structure, Host A does the same thing but flushes modified copy from cache - you get the idea.)

View solution in original post

0 Kudos
Contributor
Contributor

Thank you jhanekorn, that answer makes perfect sense!

Looks like I'll be moving forward with this project today :smileycool:

0 Kudos
Enthusiast
Enthusiast

Hi,

That was not the question. He tries to create a cluster but before starting the clustersetup first check your hardware -> common sense.

After the test the Clustersetup will not fail.

Cheers,

Jan

0 Kudos