One of the key concepts to consider when talking about services and cloud computing is the notion of coupling. We need to focus on this since, in many instances, coupling is not a good architectural choice considering that the services are not only hosted within separate data centers, but hosted by one or more cloud computing providers.
Since the beginning of computing we've been dealing with the notion of coupling, or the degree that one component is dependent upon another component, in both the domain of an application or an architecture. Lately, the movement has been toward loose coupling for some very good reasons, but many architects who build enterprise architectures that leverage cloud computing understand the motivations behind this since we don't want to become operationally dependent upon a component we don't own nor control.
Breaking this concept down to its essence, we can state that tightly coupled systems/architectures are dependent upon each other. Thus changes to any one component may prompt changes to many other components. Loosely coupled systems/architectures, in contrast, leverage independent components, and thus can operate independently. Therefore, when looking to create a SOA and leverage cloud computing resources, generally speaking, the best approach is a loosely coupled architecture.
Keep in mind, how loosely or tightly coupled your architecture exists is a matter of requirements, and not as much about what's popular. Indeed, architects need to understand the value of cloud computing and loose coupling, and make the right calls to insure that the architecture matches the business objectives. It's helpful to walk through this notion of coupling as you approach your cloud computing architecture.
With the advent of web services and SOA, we've been seeking to create architectures and systems that are more loosely coupled. Loosely coupled systems provide many advantages including support for late or dynamic binding to other components while running, and can mediate the difference in the component's structure, security model, protocols, and semantics, thus abstracting volatility.
This is in contrast to compile-time or runtime binding, which requires that you bind the components at compile time or runtime (synchronous calls), respectively, and also requires that changes be designed into all components at the same time due to the dependencies. As you can imagine, this type of coupling makes testing and component changes much more difficult, and is almost unheard of when leveraging cloud computing platforms for processes that span on-premise to the cloud providers.
The advantages of loosely coupled architectures, as found within many SOAs, are apparent to many of us who have built architectures and systems in the past, at least from a technical perspective. However, they have business value as well.
First and foremost, a loosely coupled architecture allows you to replace components, or change components, without having to make reflective changes to other components in the architecture/systems. This means businesses can change their business systems as needed, with much more agility than if the architecture/systems were more tightly coupled.
Second, developers can pick and choose the right enabling technology for the job without having to concern themselves with technical dependencies such as security models. Thus, you can build new components using a cloud-based platform, say a PaaS provider, which will work and play well with other components written in Cobol or perhaps C++which are on-premise. Same goes for persistence layers, middleware, protocols, etc., cloud delivered or on-premise. You can mix and match to exactly meet your needs, even leverage services that may exist outside of your organization without regard for how that service was created, how it communicates, nor where it is running, cloud or on-premise.
Finally, with this degree of independence, components are protected from each other and can better recover from component failure. If the cloud computing architecture is designed correctly, the failure of a single component should not take down other components in the system such as a cloud platform outage stopping the processing of key on-premise enterprise applications. Therefore, loose coupling creates architectures that are more resilient. Moreover, this also better lends itself to creating failover subsystems, moving from one instance of a component to another without affecting the other components, which is very important when using cloud computing platforms.