Exchange 2019 – Preferred Architecture


Microsoft has released their preferred architecture for Exchnage 2019 last week.

So what is Preferred Architecture – Starting with Exchange 2013 way back in April 2014, Microsoft Exchange team started releasing preferred architecture for on-premises deplopyments, based upon their experience of managing Exchange at scale in Office36. It was a set of best practices and deployment scenarios with an aim to increases availability and resiliency of those deployments.

As per Microsoft – ‘The PA is the Exchange Server Engineering Team’s best practice recommendation for what we believe is the optimum deployment architecture for Exchange Server 2019 in an on-premises environment.

Preferred Architecture is sort of presciptive in nature which makes it unsuitable for majority of customer who have some non-standard organizational requirements. However, it is best to align the architecture as close as possible to Microsoft’s PA, after all it is their product and they know it best :).

Some of the key focus areas and highlights of Exchange 2019 Preferred Architecture are:

Namespace Design – As with Exchange 2016, the recommended approach is to have unbound namespace across the site resilient datacenter pair. A quick summary on bound and unbound namespace models. With Unbound Model a single namespace is preferred, you would have 1 DAG deployed across the site resilient datacenter pair in an active-active architecture where mailboxes are dispersed across databases in the DAG. Users can be redirected by the load balancer to any of the datacenter and in the event of network (WAN) failure – neither datacenter’s connectivity is a boundary, hence the term unbound. On the other hand in the Bound Model, the users are associated or bound to one of the datacenters. There are typically 2 DAGs deployed in a active-passive architecture with certain mailboxes in one datacenter and vice versa. Here there is a need for two namespaces, one for each datacenter.

Site resilient datacenter pair design – The recommendation here is still the same as earlier with 2 or more well-connected datacenters in their own Active Directory site. This should be backed an approriate network architecture that is resilient with redundancy and low network latency.

Server Design – The recommendation is to use physical commodity servers with locally attached JBOD storage, with a scape out strategy. The new MetaCache Database needs to be on solid state drives though, which would be around 10% of overall storage needs. A quick note on MetaCache Database, it is basically the where meta information like mailbox information of a database mailbox folder structure, information about mailbox items. This MCDB helps accelerate Exchange 2019 DAG, hence needs to be on SSD with the mailbox database on JBOD.

DAG Design – There is no change in the Database Availability Group recommendations. It is same to deploy a multi-node (high number) DAG evenly distributed across site resilient datacenter pair to provide resiliency in case of a failure. Ideal to have Witness located in 3rd datacenter (could be Azure). Ensure that there are multiple database copies (2 in each dc) with 4th copy as a lagged database copy – this is the Exchange Native Data Protection without the need for backups. I am kind of proud that way back in 2013, I had architected the largest backupless Exchange for one of the largest Australian Bank, probably the first at that scale, at the point in time.

In Microsoft view – ‘By aligning with the Preferred Architecture you will take advantage of these changes and provide the best on-premises user experience possible. You will continue the tradition of having a highly reliable, predictable, and resilient Exchange deployment.

To read more about the Exchange 2019 Preferred Architecture, click here.