Recently I was asked: Is multitenancy a fundamental of cloud computing or SaaS or both? I think that depends on how loosely defined multinenancy is in your dictionary.
In computing, the term multitenancy describes a shared approach to resources. I think that in a very loose definition both cloud computing, and of course SaaS, involve resource sharing. But not all types of multitenancy are the same with respect to their approach to resource sharing.
Gartner offers up six models of multitenancy: shared-nothing; shared-hardware; shared-processing; shared-database; shared-everything and custom multitenancy.
The shared-nothing approach is the least multitenant in nature and users may have nothing more in common than a shared URL, billing system, support line or duplicated executable files. This stretches the definition of multitenancy so far as to be essentially out of bounds, in my view, and is most likely adhered to only by those seeking to follow on the bandwagon effect of SaaS and the cloud. Think of IBM WebSphere or Oracle Application Server running on Amazon EC2.
The shared-hardware approach to multi-tenancy isn’t very interesting to software types like me. In this model, the tenants share a common pool of hardware usually through virtualization. But each tenant has its own dedicated software stack (application platform, middleware, applications, databases, etc.) It should be no surprise that this is the approach taken with Microsoft Azure .NET because they want to maximize software sales and licenses and the easiest way to do that is to require multiple instances.
What Gartner calls shared-database multitenancy is about a single database for all tenants. Tenants do not share each other’s data (and can not see it), but it is stored inside the same database. However, a separate instance inside an application container is used to process the application.
Only when Gartner starts to speak about shared-processing multitenancy will some cloud advocates start to nod their heads. In this model, a multi-threaded application platform brokers each tenant’s use of the processor but each tenant will likely have a unique database instance. Gartner says: ". Most process execution resources are shared, allowing fine-grained elasticity. The application platform has multitenancy features responsible for tenant isolation and for targeting all data exchanges to the correct DBMS instances. " Gartner mentions the uniPaaS application platform as the example of this model.
In the shared-everything model of multitenancy both processor and database multitenancy is present. As Gartner suggests, this provides maximum theoretical elasticity. However, several factors such as the efficiency and design of the database gateway and the operational integrity and bi-directional scalability of the processing engine can inhibit optimal processing performance in shared-everything multitenancy approaches such as Force.com. Other downsides include public exposure of non-specific tenant metadata (statisitics about overall use of the application or platform can become known to competitors) as well as issues related to the lock-in of application code on a completely proprietary cloud platform or SaaS application.
Finally, Gartner would identify custom multitenancy as a sort of manually programmed multitenant architecture. Clearly the programming overhead is massive in such approaches and one worries about the capability of an organization to sufficiently test such an approach in a manner that can assure operational integrity and reliability.
Choosing the right level of multitenancy is an important decision. But it is not the only decision in choosing a cloud-enabled application platform. One must also consider the client environment and the challenges of developing both the server side and the client side of an application. So to answer the question we started with, yes, both cloud computing and SaaS require some model of multitenancy. If it isn't shared, it isn't cloud. Nobody owns the whole damn sky.
