Clustering Question

[imperium]

Can Windows 2003 Server R2 Edition run active/active clustering for file serving? I know that SQL is supported as active/active, but "file serving" capabilities of Windows 2003 in the past have been limited to "active/passive". This means that 1 node of the cluster is always running, while the other node sits in "standby". It is better to have active/active, because it increases performance dramatically and provides quicker failover.

Thanks for any input.

[Steven Clementi · Accepted solution]

It depends on your interpretation of "active/active"

In a Windows cluster, both nodes can have active duties. The only limitation is that they can not share the same data drives. Node 1 will always have it's own set of drives to utilize, that can failover to node 2 and node 2 will have a set of drives to utilize, that can failover to node 1. This is "active/active".

You mention SQL active/active and compare it to file serving. The fact is, BOTH are the same. An "active/active" SQL cluster simply means that you have 1 instance of SQL running on node 1 and another instance of SQL running on node 2. The 2 instances do NOT share ANY of the database files. They are on separate drives. For file serving, create 2 separate drives, 2 separate file shares in Cluster Admin and 2 separate Cluster groups each containing an IP Address resource, a Network Name resource, a Disk resource and a File Share resource. One groups is online to node 1 and the other group to node 2. "Active/Active File Sharing Cluster"

Now, if you want to be able to only have 1 file share, then you are talking about some 3rd party product since Microsoft does not support sharing physical disks between nodes of a cluster. Look to Polyserv or Veritas Storage Foundations/Cluster Suite.


The quick answer is yes, it can do active/active... in the sense that you can split your user and/or group shares to different disks and nodes. You can't do active/active with only 1 physical disk.


Steven

[Ivan Ferreira]

Windows Clusters cannot access the same disk partition at the same time, so, you cannot have both servers serving the same disk partition.

Maybe you can create a partition for every server and add a resource group for every partition, selecting a preferred server for that partition.

Another option could be the use of DFS.

[imperium]

Yes â what we need is Active/Active file serving with a shared disk. I looked a the Polyserv product, and that is what I really want to do. The N+1 all Active mode, whereby they all talk to a single data drive. My question was if Windows 2003 R2 could now do something close to what Polyserv could do. Do you know what the main difference is between Windows 2003 R2 Active/Active and their product is? I think what we want to do is Active/Active against a shared disk.

[Steven Clementi]

Windows does not allow you to share a single disk between multiple servers without your data getting corrupt.

As stated earlier, Windows Clustering does NOT support shared access to a logical drive. You need to use Polyserv OR Veritas/Symantec Storage Foundations Cluster Suite.


The difference bewteen the products is simply that. With polyserve, you can share a disk between multiple nodes. With Windows, your data with get corrupt if you try to share. You CAN'T do active/active clustering against a shared disk, MSCS will NOT allow you to even if you tried.


Steven

[Sheldon Smith]

OMG! Steven, are you saying that a SQL active/active cluster uses *duplicate* databases, one copy per host? So do the two SQL instances replicate their records to each other?
(Just when I thought Microsoft was *starting* to get a clue.... )

[Steven Clementi]

What I am saying is, you can not share physical/logical/virtual disk access between Windows Servers using only MSCS. You need some 3rd party software to "share" disk access.

As for SQL, I have only ever seen it in the way I have explained. Using that explaination, I will further say that the 2 instances are totally separate from each other... they are not for "sharing the load". Instance 1 had 1 or more databases associated with it, Instance 2 has 1 or more DIFFERENT databases associated with it. There is no replication between the 2, though i suppose you can probably set up SQL Replication between 2 DB's in different instances since they have different network names.

Active/Active, with regards to MSCS, is to utilize both nodes in a cluster, not to load balance access to the same data. Drop Polyserv in the mix or Veritas Cluster Suite and then you can share file/data access with ease.


Steven

[imperium]

Would this be possible if we used Linux as the OS?

[Steven Clementi]

Linux OS for a Windows File Server environment?


I suppose it can be done with Samba, but I am no linux expert. As well, I also do not think that linux has the inherent capability to share a logical disk and some extra software would be needed to accomplish the task, but I could be mistaken.

How much data do you have to share? Can it be split up logically?


Steven

[Ivan Ferreira]

You can use a Linux distro to enable the simultaneous access to the same data and load balance.

You need:

The Linux Distro
The Cluster Suite
The Global File System kernel module


And for this, you can chosse between:

Red Hat Enterprise + Cluster Suite + Red Hat GFS. (You must pay for all this). You get direct support from Red Hat.

CentOS, provides exactly the same (In fact, is a Red Hat Clone without copyright restrictions) but for free.

[imperium]

I checked out the Polyserve product you recommended Steve, and it does indeed provide the "missing link" we are looking for. I contacted the Polyserve and learned that they required 8 CPU minimum at a price of $52,000. At this point in time, we need only two nodes.

I would like to get more information on the other product you mentioned - Veritas Cluster Suite. On Veritas' website, the only thing I found was Veritas Cluster Suite for Solaris.

Can you point me in the right direction on Veritas' website?

[Tom O'Toole]

M$ does 'shared nothing' clustering, as others have pointed out. Using linux, you can do shared data clustering using GFS (global file system), as mentioned, another option is the oracle file system (OSFS2 I think is the current naming).

OpenVMS (and to some extent Tru64 unix), provides OS integrated shared everything clustering.

[nasdude]

Yes, it can:

http://h18006.www1.hp.com/products/storageworks/efs/index.html