Friday, September 21, 2012

SOA, Separation of concerns and agnosticism (5/5)

Separation of concerns and abstraction

Any service that we build should have a certain level of abstraction. Abstraction comes in many kinds but they are all intended to provide separation of concerns.

As an example. if a service accesses a database and the database has an VARCHAR2() field for a numeric value called DBLADR2FLD, then it should not matter to the service consumer what the underlying datatype and column name of the database field is. In essence the service abstracts from the implementation to prevent this.

Additionally, a VARCHAR2() datatype is a specific vendor technology datatype. The service should also abstract from the specific database vendor technology.

Why we do this can be answered pretty simple and straight-forward: the underlying data model and the database vendor technology can change over time or be replaced by a completely different system or technology, and this should not matter to the consumer of the service. This is the most important reason why we have abstraction. By doing contract-first approach, we can already make sure that up-front no implementation-specific or technology-specific dependency can trickle into the contract and as a consequence into the service consumer logic (see [3])

Another form of abstraction is the quality of service abstraction: not all quality of service information does need to be known to the consumer of a service. Important QoS attributes are the availability and the reliability of a service, how much load it can handle and what the average and peak duration of execution is.

Not important however is the fact that the service had 4 failures in the past hour, as this does not say anything about how it will behave in the future and also it might even make the service look unattractive. In a sense we need to be shielding these kind of runtime concerns from potential service consumers. This can be done by creating SLAs that cover the quality of service in a constructive way with meaningful performance indicators.

Many more types of abstraction can be applied - if I can find some more time I will update this article.

- Roger

No comments:

Post a Comment