Hi All.

I have a few questions regarding a deployment that we are planning to do in a new installation. Currently we are running a 4 node Hyper-V 2012 cluster connected to iSCSI storage. The general details are:

Nodes in cluster: 4 / Physical OS: Windows 2012 Datacenter Edition / Storage: iSCSI

iSCSI Storage Network: 172.128.x.x (isolated, not routed)

Production Network LAN: 10.252.x.x

Amount of mailboxes to create: 1000

Exchange Server version: 2013

Virtual Machines: All data is stored on the iSCSI SAN. No local storage can/will be used.

VMs Running Exchange 2013 Mailbox Role (desired): One in blade1 and One in blade2

VMs Running CAS role(desired): One in blade3 and one in blade4

Goals:

1- To balance the load of the users in two MBX VMs (500 mailboxes each)

2- To use 2 CAS VMs to balance load

3- To protect each MBX VM databases in case of failure (with a database copy??)

4- (future) To create a replica/failover on another physical distant site.

Questions:

Should I use DAGs even if I am using a cluster and all data will reside on the iSCSI SAN?

What kind of recommendations exist for such scenarios?

Ive only seen documents talking about hyperv2008 and exchange 2010 and since there are so many changes in both products, I better ask before i begin setup. 

Thanks in advance.

Regarding Goal 1, why bother?  The passive mailbox server works almost as hard as the active server, and neither should be heavily taxed with just 500 users.

Goal 2:  Network or hardware load balancing.

Goal 3:  Multiple DAG copies.

Goal 4:  Multiple DAG copies.  However, unless you have a rock-solid, fast WAN, I wouldn't advise you to expect high availability between sites.  But you can have site resilience with a bit of downtime for you to manually switch over.  Myself, I wouldn't ever want Exchange to automatically fail over to a remote site, and it's my experience that Exchange will do just that when you least expect it and it's hard or even impossible to figure out why.

Do not use any virtual high-availability solutions with Exchange.  Do not allow Hyper-V or VMware move Exchange VMs around for you.  In general, it is not supported and you won't like the results.  The DAG works well and gives you everything you need.

Hi Ed, thanks for the fast response.

However you mention "the passive mailbox". We are planning to have 1000 users and the whole idea of having two MBX virtual machines is to split 500 mailboxes on each server and give the user the best experience possible in a virtualized environment. Unless of course I am completely wrong.

Also, are multiple CAS VMs ok?

In case your expertise recommens a DAG, I feel Im clustering the cluster you know? ;)

And if DAGs are the way to go, I will need 3 interfaces per VM, right? MAPI, Replication and the user-exposed interface? and the MAPI and Replication interfaces must be routable right?

Can you please clarify?

500, 1000, it doesn't matter.  A mailbox server works nearly as hard handling passive mailbox databases as active ones.  Besides, you have to design your mailbox server so that one can handle all the load anyway.  That's what fault tolerance is all about.

I don't know what you're asking about "multiple CAS VMs ok".

I explained my and Microsoft's position about virtual fault tolerance for Exchange servers.  If you think you're smarter than me and Microsoft, then go ahead, but don't come back here crying when you shoot yourself in the foot by corrupting your databases.

There is no requirement for multiple interfaces, but a second interface is strongly recommended.  The MAPI network is the user interface; I choose to call it the public network.  The second "private" network is there mainly for the good old Windows Clustering reason, to provide an alternate path for the heartbeat in order to reduce false failovers.  Since it is there, you can use it for replication, but you can just as well replicate over the MAPI network.  In fact, if you configure a replication network, and it goes down, replication will switch to the MAPI network.  The private network is not generally routable, but if you want to join private networks between datacenters, you might choose to join them via a second WAN link, in which case you'd configure static routing.  But joining replication networks between datacenters is not required, and I don't advise it unless a true redundant network path is available.

Ed,

Regarding the "multiple CAS role virtual machines" I was referring to a CAS-array. In order not to have a single CAS role server.

About the mailbox server, indeed, one should always plan for a complete server load.

And for the "think you're smarter than me and Microsoft"... I definitely dont think so, otherwise I wont be asking here in the forums. I was just asking for advice on this scenario in order not to make a mistake. That is the whole idea of asking kind-of-dumb questions expecting a guided answer.

Logic perhaps, is different than a technnical design. Because human logic dictates that two servers can perform better than one. However I found several documents that consider my user count as "small" and one server will probably be (more than) enough.

Thanks for the clarification of the MAPI and Replication network interfaces. I had the wrong idea that I needed a third interface for "user interface".

The CAS array doesn't really exist in Exchange 2013 anymore.  You just have two or more load-balanced client access servers for client access high availability since they just take care or protocol routing and there's no longer support for MAPI RPC.

Good luck with your deplo