We are planning on upgrading from Exchange 2010 to 2013.  

Currently, we have 2 Exchange servers.  One server with mailbox and hub transport in our local office with users.  The CAS is in a remote data center.  So, whenever either server is down or being restarted, all mail stops.

We have received licenses for three Exchange 2013 servers.  What is the best way to set these up so there is high availability whenever any one of the three servers is offline.  Also, since we will still have the licenses for our existing two Exchange 2010 servers, will there be any benefit in keeping those in this scenario?

I have attached a diagram of our current layout and we want to revamp it.

Hi 

The current setup of your environment is not Microsoft recommended solution as there is no HA set up for mailbox hub as well as CAS 

My recommendation for exchange 2013

You can have CAS servers deployed one in each datacenter 

Below thing can be thought for DAG Config

In order to increase the service level which is to have one site alive when the other site is down, its better to Move the file share witness server to a server that does not belong to neither site but it needs to have reliable network communications between both sites.

So if any of two sites fails, the DAG will have a quorum and it can remain operational. The databases will not be dismounted and service and data access will not be disrupted.

If in case if you don't have 3 sites you can follow below( again this is not best recommended HA )

Put the FSW in the primary Data Center and set the DAG to DAC mode so you can easily do a data center switchover in case the primary fails.

http://technet.microsoft.com/en-us/library/dd351049(v=exchg.150).aspx

Couldn't our local office be considered a 3rd site and we just put the witness server in the building with end users instead in any of the data centers?

So, if we put one Exchange server in each of the 2 data centers, can we also put the third Exchange server in our local office so we can send emails to other users in the building without the user's computers having to communicate with either of the data centers across the WAN connection, have snappier local email performance and also save bandwidth?

The way the email is set up now, when the user's are working in the office during business hours, their computers always have a fast LAN connection to their mailboxes which might make Outlook respond faster and attachments download much quicker than if they always had to communicate across the WAN to one of the remote data centers to access their mail.

• Edited by MyGposts Sunday, February 08, 2015 6:54 AM

Couldn't our local office be considered a 3rd site and we just put the witness server in the building with end users instead in any of the data centers?

So, if we put one Exchange server in each of the 2 data centers, can we also put the third Exchange server in our local office so we can send emails to other users in the building without the user's computers having to communicate with either of the data centers across the WAN connection, have snappier local email performance and also save bandwidth?

The way the email is set up now, when the user's are working in the office during business hours, their computers always have a fast LAN connection to their mailboxes which might make Outlook respond faster and attachments download much quicker than if they always had to communicate across the WAN to one of the remote data centers to access their mail.

• Edited by MyGposts Sunday, February 08, 2015 6:54 AM

Do you want copies of your data on two servers or three? You say "any one of the three servers is offline" so I'll assume two. My first thought would be to use just two multi-role servers, one in each datacenter. Then you would surely put the FSW in your local office, along with an unbound namespace to the two datacenters.

With a third server in the mix, you could go one of two ways. The first is to put a second CAS server in your primary datacenter (if you want three copies of your data, make it multi-role). Put a hardware load balancer on top of those two servers, and remove the secondary datacenter from your namespace. The advantage is that your users will not be directed to your secondary datacenter, which may have higher latency. The disadvantage is that if your primary datacenter fails, you will have an outage until you change the IP address of your namespace. And the cost of the load balancer.

My other idea is to put the third CAS server in your local office. Bringing the CAS close to your users has a big effect on latency. The downside here is how much harder the namespace planning becomes. If you used anycast to your three CAS servers your in-office users would hit the local CAS, while your in-field users would hit their closest datacenter. I think the difficult part would be removing routes when the CAS fails. Without this happening automatically, some of your users will have an outage when a CAS fails.

Since we have already bought and paid for 3 Exchange 2013 licenses, we might as well take advantage of all of them.  Maybe we could even buy a 4th it would make a huge difference in this layout.

We would like to be able to reboot any one server at any time without it creating an outage.  Is that possible to do with 3 servers?

What about having 3 multi-role servers?  One in the local office and one in each data center?  We do have many users that work remotely, but the vast majority of mail traffic is in the local office during business hours and maybe having a multi-role Exchange server with both CAS and mailbox roles on the local LAN will make Outlook snappy locally.

Is there any way to design this so that, if we add one of these Exchange servers to our office LAN and it goes down or is in process of rebooting, that internal office users would seamlessly get redirected to a server in the remote data center and then start using the local office server again once it is fully online?

Do you want copies of your data on two servers or three? You say "any one of the three servers is offline" so I'll assume two. My first thought would be to use just two multi-role servers, one in each datacenter. Then you would surely put the FSW in your local office, along with an unbound namespace to the two datacenters.

With a third server in the mix, you could go one of two ways. The first is to put a second CAS server in your primary datacenter (if you want three copies of your data, make it multi-role). Put a hardware load balancer on top of those two servers, and remove the secondary datacenter from your namespace. The advantage is that your users will not be directed to your secondary datacenter, which may have higher latency. The disadvantage is that if your primary datacenter fails, you will have an outage until you change the IP address of your namespace. And the cost of the load balancer.

My other idea is to put the third CAS server in your local office. Bringing the CAS close to your users has a big effect on latency. The downside here is how much harder the namespace planning becomes. If you used anycast to your three CAS servers your in-office users would hit the local CAS, while your in-field users would hit their closest datacenter. I think the difficult part would be removing routes when the CAS fails. Without this happening automatically, some of your users will have an outage when a CAS fails.

• Marked as answer by MyGposts 11 hours 6 minutes ago

If your DNS provider supports geo-DNS, then it could return the local office CAS first, followed by the other two, for the users in the office. If that server fails, clients would retry to one of the other two CAS servers. This would send all of the local office load to one server in the normal case, so the hardware requirements would be quite high. Also consider that your local office may not be as reliable a host as your datacenters.

I think the most common deployment in this circumstance is even numbers of multi-role servers in your two datacenters, with the FSW in your local office.

We use Windows Server for DNS, so I don't think geo-DNS is part of that.

So unless we get a 4th Exchange server, it sounds like the best we can do is put 2 multi-role servers and a load balancer in the primary data center and the third multi-role in the secondary data center that doesn't get used as a CAS unless the first data center is offline.  To make this work if there is a break in communication between data centers and our offices, we would probably also need domain controllers and DNS local to each data center.

Can we make a 3 member DAG out of this design with two servers in the primary data center, the third in the secondary data center and the FSW in the local office?

With that layout,  shouldn't we be able to reboot any of the servers individually without an outage?  I think the only outage that could happen would be if multiple servers went down at the same time or the primary datacenter was inaccessible and IP addresses had to be updated to point to single server the secondary data center.

The disadvantage is that we would not have any Exchange server that is local on the LAN and that could lead to sluggish Outlook performance. 

Yes, that would work just fine. There won't be an outage with any server failure. The downside is that there will be an outage if your primary datacenter fails, until you go through your site activation process.

There's little advantage to putting the FSW in a third location if you have an uneven number of mailbox servers.

Don't worry about latency very much. If it's less than about 100ms, OWA will be completely fine. Outlook desktop is much more tolerant than that.

OK sounds good and we will skip the FSW since there would be 3 members of the DAG.