Clustering

With a major upgrade imminent I've been looking into the possibility of clustering as opposed to NLB. I know linux (Beowolf?) can and does it easily enough but of course we'll be running Windows here (My choice really ;)).

A few questions to the community at large.

Firstly, I believe the clustering option to only be available to the Enterprise or Datacentre edition of Windows 2003 R2 or have I been misinformed?

Secondly, has anyone else set this up and seen an improvement in comparison to what existed before?

Thirdly, having looked in to large storage solutions I would look to put in place two large NAS boxes and setup DFS, can anyone see a flaw in this?

I want to create the most reliable network possible (Within the budget limits of course! ;)) I would have thought that either NLB or Clustering would give me good transparency for Authentication and Authorisation etc and DFS good transparency for file storage e.g. User Areas, Departmental Learning Resources etc.. Having said this does anyone know of a good way to do the same with Printers/Printer Queues?

Wes

Quote:

---

Firstly, I believe the clustering option to only be available to the Enterprise or Datacentre edition of Windows 2003 R2

---

Correct.

Quote:

---

Secondly, has anyone else set this up and seen an improvement in comparison to what existed before?

---

Not with windows, but it depends what applications your running. Exchange, SQL Server, IIS support clustering for example.

Quote:

---

Thirdly, having looked in to large storage solutions I would look to put in place two large NAS boxes and setup DFS, can anyone see a flaw in this?

---

So whats the clustering for then? I've found DFS to be unreliable.

Quote:

---

Having said this does anyone know of a good way to do the same with Printers/Printer Queues

---

Printer pooling? Use a Linux box with failover?

I was looking at the clustering for just DC reliability, but if you think that may be a little OTT then NLB would be the next step. Printer Pooling using a linux box? Is there a printer credit system available for Linux that can be integrated with AD as our school uses credits extensively (we charge for excessive printing)?

I'm looking to install a couple of DCs but they wouldn't store anything else other than the OS and AD database. The storage would be done seperately I don't want to put too much on any server (All eggs in one basket)

What would you recommend Geoff?

Wes

Quote:

---

Printer Pooling using a linux box? Is there a printer credit system available for Linux that can be integrated with AD as our school uses credits extensively (we charge for excessive printing)?

---

CUPS + Samba.

Quote:

---

I'm looking to install a couple of DCs but they wouldn't store anything else other than the OS and AD database. The storage would be done seperately I don't want to put too much on any server (All eggs in one basket)

What would you recommend Geoff?

---

Just get a few basic 1U rack mounted servers. Set them up as DC's and scatter them around your site.

I dont think you can cluster DCs as such. Something I read recently was saying about clustering and replication and how that other DCs in a cluster would not be upto date as they only come 'live' when the other goes down.
It's something along those lines anyway :D

Is that other DCs or other nodes within the cluster though? If you have say 4 servers/nodes in the cluster then shouldn't they all be upto date?

Each works as if part of the whole?

Wes

You can't cluster DCs. It's non-sensical.

I thought a cluster was a group of servers setup so they acted as one? So you could have say 2 clusters one entity as a DC another as a storage server?

Wes

A fileserver does not have to be a DC.

DC's already have redundancy built in due to the way active directory works. There's no need to add another layer of redundancy/complexity by clustering them. Indeed, it isn't a supported configuration.

Fair enough judging from the extra cost of Enterprise it may not be a financially sound idea either. The setup I'm looking at currently wouldn't have the DCs being fileservers anyway I'd rather have everything distributed across several servers.

Wes

Quote:

---

Originally Posted by Geoff

A fileserver does not have to be a DC.

DC's already have redundancy built in due to the way active directory works. There's no need to add another layer of redundancy/complexity by clustering them. Indeed, it isn't a supported configuration.

---

The only thing that may not have a smooth failover is FSMO. I haven't tried but can roles be seized post mortem, as it were?

Has any onr tried using FRS to create redundant file storage?

You can but I think the Schema master maybe the only one to give you trouble in that area. If the original FSMO master has been stolen, corrupted or otherwise unavailable then you need NTDSUTIL (Info here: KB255504).

Microsoft spends most of its time saying that DFS is much better than FRS but after Geoff saying DFS is unreliable I would have thought FRS would be a more stable alernative without too much trouble setting it up, hadn't thought about it before, thanks!

Wes

NTDSUTIL is fairly easy to use to seize or transfer all roles. Remember not all roles can be transfered through the GUI.

I believe the command line options can do that though ChrisH? Or am I incorrect?

Wes

Thats what I meant in my first sentence by 'all roles' :wink: