Hey guys,

Exchange 2013 design setup question.  My customer has 3 datacenters and wants to setup one multi-role Exchange 2013 server in each.  They have 100 mbps connection between them.  For client access I was going to setup round robin DNS so clients can connect to any of the 3 servers.  For the DAG, I was going to stretch is across all 3 servers and have a copy of each database on each server.  Please let me know your thoughts on this design and if I am missing anything.

My confusion is with quorum and the FSW.  I want to make sure I have this right.  With a 3 node DAG, I need two votes for it to stay online.  So if I lose 2 sites, will the DAG stay online as long as one server is online and the FSW is reachable?

Thank you for your help.

Kevin

I would say this is a rare design. But I don't see any issue out of my mind, at least for Outlook and OWA running on Windows client. DAG across 3 data center works for sure. 3 CAS in 3 data center with the same name should work as well.

But I would strongly recommend to have a comprehensive lab test before going to production. Anyway, it's rare design, as I mentioned.

As for the quorum, you can forget about the witeness. A DAG with odd number of node does not use the witness at all. If you lose 2 nodes, you lose the quorum.

• Proposed as answer by Allen_WangJFMicrosoft contingent staff, Moderator 1 hour 47 minutes ago

If you lose 2 sites at the same time, the answer is ultimately no, you would need to perform a dataceter swithover.  If you lose 1 site at a time you should be able to use dynamic quorum to reconfigure the cluster dynamically.

http://blogs.technet.com/b/scottschnoll/archive/2014/02/25/database-availability-groups-and-windows-server-2012-r2.aspx

If it was me and I had 3 sites, I would pick a site that I would call my "primary" datacenter and depending on the number of users i have I would do one of three things:

1. If I don't care about automatic failover if i use a site, I would setup 2+1 DAG where I have 2 nodes in my primary site, and 1 node in what i would call a DR site.  
2. If I want auto failover then I would create a 2 +2 DAG with 2 nodes in a primary site and 2 nodes in one of the other 2 sites and use the 3rd site as my file share witness.
3. If my requirement is to keep users mailboxes on a server in their local site, then I would probably create an environment that is a bit overkill, but builds in local site HA for each site and create 1 dag stretched over the 3 sites, with 2 nodes in each site so I can lose a server and keep mail services for that site in that site.  I would pick a main site (or possibily use Azure) to put the FSW in.  Keep in mind if the WAN Link for all sites go down then the two Exchange servers in the sites that do not have the file share witness will dismount the databases and (assuming each server has a copy of all the databases) your mail will be running out of the site with the fileshare witness.

I would say this is a rare design. But I don't see any issue out of my mind, at least for Outlook and OWA running on Windows client. DAG across 3 data center works for sure. 3 CAS in 3 data center with the same name should work as well.

But I would strongly recommend to have a comprehensive lab test before going to production. Anyway, it's rare design, as I mentioned.

As for the quorum, you can forget about the witeness. A DAG with odd number of node does not use the witness at all. If you lose 2 nodes, you lose the quorum.

• Proposed as answer by Allen_WangJFMicrosoft contingent staff, Moderator Friday, August 28, 2015 5:39 AM
• Marked as answer by BestBet 17 hours 57 minutes ago

I would say this is a rare design. But I don't see any issue out of my mind, at least for Outlook and OWA running on Windows client. DAG across 3 data center works for sure. 3 CAS in 3 data center with the same name should work as well.

But I would strongly recommend to have a comprehensive lab test before going to production. Anyway, it's rare design, as I mentioned.

As for the quorum, you can forget about the witeness. A DAG with odd number of node does not use the witness at all. If you lose 2 nodes, you lose the quorum.

• Proposed as answer by Allen_WangJFMicrosoft contingent staff, Moderator Friday, August 28, 2015 5:39 AM
• Marked as answer by BestBet Friday, August 28, 2015 1:33 PM

Thanks for the reply Li.  This was just what  I was looking for.

That's *IF* you lose DCs one by one. A WAN outage/flap would take the entire DAG offline in a single go.

Dynamic quorum does not always save you....

This is not a design that would not be recommended.  It is not in the preferred architecture.

Hi Rhoderick,

In this scenario, of the three datacenters, there are users in two of the datacenters.  The third datacenter is used for disaster recovery for several other applications.  There are separate internet connections into each datacenter and 100 mbps MPLS connections between all of three sites.  The customer, before asking for help, already had purchased three servers, one for each datacenter.

I took that information and was asking here in this forum, if we should go with an unbound model that sends traffic to all three sites, I would setup MX records to send mail to all three sites as well.  And a DAG that stretches across the three sites.

What design would you suggest here?  I am open to all suggestions.  I want to give them the best availability with site resiliency that we can.  The preferred architecture does not exactly cover this scenario where there are three active datacenters we can use.

Thank you.

Kevin