Recommended hardware structure for Microsoft exchange server.

The present way of thinking is a high-end server with multiple drives, raid configuration, redundant power supply, network cards and…

But that is no longer the case. 

With exchange 2016 Microsoft recommends multiple less expensive servers with regular SATA hard drive configured for JBOD.

IMPORTANT: you should have Raid (can be Raid 1) for the application drives as Exchange will only handle the database replication.

Why is this a better model?

First Money :

1. Think how much money you save just on the hard drives, 1- by getting get full space of each drive which you would not get on Raid 10 or 5 so you need fewer drivers, 2- by using SATA drives instead of SAN/SAS/SSD.
2. On the server, you no longer need this expensive server since you have multiple servers handling the load instead as when one server handling the entire load. Same is with componence for redundancy with having had multiple servers even without redundancy on the server itself you get redundancy by having more than one server
    

Second Reliability :

1. There are some components in a server that if they fail, the server will fail since there is no redundancy for it. For example the raid controller, the mother board. With using multiple machines there is no component that if it fails it will stop the server from running. (And of course, multiple VM's on one physical machine will not accomplish the goal).
2. By having exchange doing the replication instead of the Raid controller, Raid controller is hardware replication 1 and 0 but replication that is done on the application level is smarter by definition as its application aware, so for example will only replicate the transaction logs and not the entire data, which can make a big difference to performance. Just have hard drive designated for hot spare and when exchange detects that an active disk fails, it will invoke the hot spare, format it and copy the data to bring it into operation.

What is new in exchange 2016 that enable this model?

1. Exchange 2016 has built in advanced failover system.
2. It reduced the amount IOPS it uses to 10 IOPS (a SATA drive gives you 70 IOPS)

One very important consideration before you implement this change is that you still need a Raid controller because of the Cache, which is critical for the performance of the exchange server, and make sure to enable the cache policy to 100% Write cache

The content of this article is discussed in depth in Microsoft channel on YouTube (Microsoft Ignite) and can be found at the following link https://youtu.be/z32lmrkfanU 

Since the purpose of my writing this article is only to make you aware of the big change in the new version of Exchange, I've only briefly discussed what I found to be the main points. You should, therefore, read the Microsoft documentation available at the provided links below..

• DAG: Beyond the “A”
• The Preferred Architecture
• The Exchange 2016 Preferred Architecture
• “Partitioned” Cluster Networks
• Designing a Highly Available Database Copy Layout
• Ask the Perf Guy: How big is too BIG?