As with every product VMware delivers, we make various changes and updates to the back end services to deliver a best-in-class product to all our customers. Certain products or functionality are sunsetted and other products or functionality are enhanced or newly delivered.
AirWatch Cloud Messaging (AWCM) clustering is one of the more confusing and difficult pieces to implement. Which clustering should I use, or which can I use? Do I need clustering? How do I configure clustering?
Let's Start with the Basics
Implicit Clustering with AWCM: Implicit clustering is truly not clustering at all - it relies on load balancer configurations to ensure that clients are delivered to the same AWCM servers that their requests first connected to. This keys off an application-based cookie that the client sends in the request header or URI and the load balancer uses to determine which server to persist to, which can make for some complicated load balancing.
Explicit Clustering with AWCM: Explicit clustering is clustering in the truest sense of the word. Underlying communication occurs between all nodes to ensure that connections are routed to the correct AWCM server. Complex load balancing is not required and there is no need to consider the application-based cookie because it doesn’t matter which server you land on; you will always get forwarded to the correct place.
VMware Development Direction
VMware continues to support both implicit and explicit clustering. However, from a development perspective, explicit clustering is a feature that is in maintenance mode only - current development work and improvements are not being applied to explicit clustering as they are for implicit clustering.
You might have noticed the slow removal of references to explicit clustering in most documentation. This was done specifically to curb the implementations of explicit clustering for some of the reasons that will be covered later in this post.
As such, the default recommendation that customers should put in place is implicit clustering for its performance benefits and for its longevity within the VMware Workspace ONE UEM ecosystem.
Default Recommendation Implicit Clustering
VMware has recommendations on how to enable this configuration for F5 and NetScaler load balancers in the KB: Setting up AWCM persistence rules using F5 and Netscaler (2960904). If the load balancer has the ability to persist on that application-based cookie, persistence will function as intended for AWCM implicit persistence.
Additionally, we are working with various vendors to add supportability for AWCM style load balancing requirements to ease the complex persistence requirement. There is no detail on when these might be included but continue to be on the lookout for announcements or modifications to the KB with additional configuration options.
Explicit Clustering Limitations and Thresholds
If AWCM implicit cannot be implemented, explicit is an option for some customers. Example scenarios include: if the load balancer or configuration does not support AWCM implicit or if there is a network requirement that might prevent its implementation (for example, SSL offloading is not permitted). If the deployment has one of these requirements and is under 20,000 total devices, AWCM explicit could be implemented as an alternative.
A configuration of implicit with active/passive could still be considered instead of implementing explicit but would require a well-vetted load balancing policy. This implicit configuration limitation would become an issue at around 35,000 - 40,000 concurrent connections (devices) and we would be forced to implement active/active implicit clustering.
Future for Explicit Clustering
As we have stated, explicit clustering is in maintenance mode with respect to development and feature implementation. We are looking into explicit clustering for future iterations of Workspace ONE but currently, significant performance improvements to explicit would be required to ensure that sizing does not significantly change for customers. There is also no official sunset or EOL for explicit clustering, but it is important to understand the previously covered limitations.
We hope to have more information about the possible resurgence of explicit clustering as we continue to work through some key projects related to deployment architecture.