Microservices. Service-oriented, but not SOA, this architecture is becoming more common as mobility and time to market drives up the ante in this high stakes game of applications.

But just what are microservices? If you want a more complete and lengthy view I highly recommend reading Martin Fowler's series on the topic. If you're looking for a summary, then read on.

Microservices are the result of decomposing applications. That may sound a lot like SOA, but SOA was based on an object-oriented (noun) premise; that is, services were built around an object - like a customer - with all the necessary operations (functions) that go along with it. SOA was also founded on a variety of standards (most of them coming out of OASIS) like SOAP, WSDL, XML and UDDI. Microservices have no standards (at least none deriving from a standards body or organization) and can be based on nouns, like a customer or a product, but just as easily can be based on verbs; that is, functional groups of actions that users take when interacting with an application like "login" or "checkout."

Microservices changes the scalability model of an application. A traditional, monolithic application is typically scaled out, horizontally, using an x-axis scaling pattern. That means as demand increases, additional copies (clones) of the application are added to a pool. A load balancing service provided by a proxy or an application delivery controller (ADC) distribute requests across that pool.

Microservices can still be scaled out horizontally, but across individual services. In other words, microservices breaks down applications into multiple components based on functional or object groupings. Each service is individually scaled, with the result being that the application is scaled more efficiently.That Monday morning rush to log in to the app winds up forcing a scaling event for the "login" service, but not the "logout" or "search" service.

What this means is that operations is prone to isolate each service in its own scalability domain, each with its own dedicated control point (proxy or ADC). The more services, the more control points you need.

There are two ways to do this (well, probably more but for the sake of brevity let's focus on the main two ways). One is to use a very high capacity ADC capable of acting as a central control point for hundreds (or thousands) of services. The other is to distribute that control to each service, essentially building out a tree-like hierarchy of control points at which services can be scaled.

It is the latter that is increasingly popular and the architectural scalability pattern for applications is becoming one that scales the application with a central command and control ADC while scaling out its composite services using service-specific proxies.

This makes sense from the perspective of adjusting application behavior to ensure performance and security of applications based on context (the unique combination of device, network and application) while enabling available and scale of the services that comprise that application. That's because those same services can (and many would argue, should) be used in other applications. Tuning a service for one application no longer makes sense, as tweaking TCP and HTTP options for one application can actually be detrimental to another. By delegating responsibility for app performance and security to the "thing" that virtualizes the app - the ADC - and enabling the "thing" responsible for scaling a specific service - the proxy - the result is a finer grained scaling architecture that is better able to adapt to service-specific and app specific requirements.

All this is well and good, but how does that drive operationalization?

Driving the need to Operationalize 

Well, if you consider how many services might need scaling, you can start seeing that it's impractical to manually manage the process. With hundreds (or more) services needing scalability on-demand, manual processes that include launching a new instance of the service and adding it to the proxy pool (and reversing that process when demand diminishes) is simply not feasible.

Operationalization - people collaborating and using programmability to optimize the processes necessary to meet business priorities - takes the approach that these processes, which can be abstracted and encapsulated into a well-defined set of steps, should be orchestrated via programmability (APIs) based on jointly determined (between Dev, Ops, and the Business) thresholds. Those thresholds might be based on performance or capacity or both; the important thing is to ensure they're determined collaboratively to ensure the overall application is able to meet the expectations of the business (and users).

If an application that typically needed one set of services is decomposed into ten different services, then ostensibly you'll need 10 times the services. Which is a lot of services. Services that are, because they're deployed in the network between users and apps, managed by network and operations staff.

load carryThe reality is you can't grow staff at the rate required to handle the load. But it's also true that it isn't the load that breaks you down, it's how you carry it.

Thus, operationalizing the provisioning, configuration, and lifecycle management of those services using programmability (APIs, automation and orchestration frameworks) can dramatically reduce the operational impact on this sudden explosion of services. 

As we saw with server virtualization, templating and automated provisioning and management enable server admins to scale from managing fewer servers to averages that would be impossible to match with physical counterparts. The same thing has to happen in the network if IT is going to support the explosion of app services needed to maintain the performance, security and scale of microservices.

That doesn't necessarily mean the network has to go virtual, but it does need to enable the same characteristics as that of its more automated, virtual counterparts in the compute domain. That is, they need to be rapidly provisioned and API-enabled and support templatization to provide the means by which they can be centrally managed through automation and orchestration.

We really need to operationalize all the network things.