Wednesday, March 31, 2010

Meetings kill Loose Coupling

There is a brilliant, yet underutilized, sociological observation called the Conway’s Law. It was stated by Mel Conway in his 1968 paper. It states that:

Any organization that designs a system will inevitably produce a design whose structure is a copy of the organization's communication structure.

The key phrase is “communication structure”, that is roughly, but not always, equal to organization structure.
You may have not heard of or read his paper, but I am sure you have seen the effect of his observation:

If you have two teams called Application and Kernel, then chances are you end up with two deployment units called applications and kernel (jars or folders or zip files or any other form of bundling code), on the other hand if you engineering organization has two teams called Books and Games, then you end up with a book and a game application, if there is no other team/group, it is unlikely that you can see a kernel or core subsystem that encapsulate the common construct between the two.

Now, we all know about the principal of loose coupling, and how it enables flexibility, efficiency, increased productivity etc. One very effective way of creating loose coupling between two system (let’s call then consumer and producer) is to make sure that members of the producer team do not meet with members of consumer team! You may convey requirements to them, but forbid meetings. If there is not communication, it is hard to build hard coupling. It is a bit strange and counter-intuitive, but we have used this with success in building key infrastructure services.

I have found out that a good early predictor of the level of coupling (or quality of interfaces in general) of a system is the list of invitees to their early design meetings. Whenever one of few engineers from a future consumer is part of the meeting, there is a good indication that they systems will be somehow coupled (either thru the types or domain values of parameters passed or the way errors are handled, the marshalling of the result, or even the terminology used) – This of course is a generalization, the team somehow has to collect requirements, or share result s and get early feedback, this is all OK, but if your goal is to create “loosely coupled” system, you should make sure your communication structure,  are loosely coupled as well.
Take a service that verifies a credit card number to ensure validity and that a provided name indeed matches credit card issuer record. This service may have a method/operation in its interface:

Result VerifyCreditCard(UserId id)

It assumes that somehow the service can obtain a credit card number from a supplied UserID. This is a very tightly coupled interface that shows the service provider has had too much knowledge about its consumer.
Here is a less tightly coupled (better) example:

Result VerifyCreditCard(CreditCard card, BuyerName buyerName)

This is not too bad, but the choice of “BuyerName” indicates that the provider has the knowledge that consumer (the one she probably met with) happens to deal with principals that are buyer.
Consider a loosely coupled version that one would probably write if there is no additional knowledge of potential consumers

Result VerifyCreditCard(CreditCard card, Name name)

Communication channels are very important interface design, this means people in each design meetings should be selected carefully, if system A should not be dependent on system B, the best way to ensure it, is to reduce communication between developers of system A and system B.

Tuesday, March 30, 2010

Evolving Gracefully - Example I

In my posts, "What is Architecture", Part I and Part II, I argued that architecture is essentially what enables a system to evolve gracefully and traverse an optimal path in response to changes (requirements, assumptions, context, inputs etc.). In this post (and in a few more) I give examples of such changes and possible optimal/graceful (or non-optimal/disruptive) responses of a system.

Take "Accessibility Requirement" as an example. Few software systems are designed with the explicit goal of accessibility as one of the main architecture and design goals. However, some systems at some point must comply to certain accessibility standards and guidelines (it is good usability practices, the right thing to do, good business and the law) - see a very good introduction to accessibility from WebAIM.org here.

Now, if some one came one morning and asked you to make sure every image in every single HTML page of your site has an alt tag, what would be your response? Remember this maybe 1000s or image tag in 10s or 100s of applications produced by developers across three continents.

If your architecture (and the design and implementation of it) decoupled model construction from actual rendering, and a centralized rendering engine that takes a data structure (such as DOM) and uses a rendering strategy to produce XHTML, then your architecture (and you) has no problem responding to this new requirement easily. You'd simply make sure that when the render visit an element, it finds an alt attribute. This is an optimal response to change. Although the system was not particullary designed with accessibility in mind, since it was built based on the right architecture, it can evolve to accommodate change.

However, if the 1000s HTML pages on your site are produced by 100s of JSP scripts, a few PHP scripts, some XSL, a few instances of dynamically created tags in JavsScript etc. You would have a nightmare on your hand. This is an example of architecture that can not evolve ...

Monday, March 29, 2010

Identity Assertion Framework (IAF) Talk at Stanford

 I will give a talk about eBay's framework for handling federation, federated token services and distributed authentication - called IAF - at Stanford on Monday April 19th, 2010. Here is a brief description  The talk is part of the Stanford Security Seminar.

This is an area that is poorly understood by software developer and architects that are new to security or authentication. I will write more about the topic of "authentication" and eBay's framework for handling it in more details. Look for under the label "Identity" and "Security".

Friday, March 26, 2010

What is Software Architecture - Part II

In part I, I offered my view of software architecture by illustrating what "architecture" does for you: Allows you to deal with change optimally. This is basically saying architecture is what allows a system to change with out changing fundamentally what it is (forgetting about Dialectics for a moment). In other words:

  Architecture is what enables a system to evolve gracefully

You may ask, how else a system would change? How is this for an example? Don't let the glitz and glitter fool you. This is an example of architecture that fail to evolve according to the changes in its context and accumulation of demands for change finally resulted in its destruction. However, I doubt that you see an implosion of Taj Mahal - architect Ustad (Master) Ahmad Lahouri - any time soon.

Now you may say, well these are building, what do they have to do with software? I don't want to over play the building metaphor, but software systems acts similarly. How many times you have seen a multi million dollar investment in, say, customer support system or CRM system or messaging infrastructure or authorization system etc. only to scrap them two, or three years later by another multi million dollar system?

On the start up side, often the first and only concern is to make it work and to survive, but when time comes to scale, they often have to re-write from scratch, that is the software counterpart of a Vegas hotel implosion.

In my next post, I will give you a few software and system examples of graceful evolution courtesy of good architecture practice and "implosion" for lack thereof.

Thursday, March 25, 2010

What is Software Architecture - Part I

My job title carries the word “architect” - although my Mom never fails to point out that I am not a real architect like my brother is. This title among the software engineering community in general and in the Silicon Valley in particular almost always is received as an incomplete description of what a person does. People always expect an additional clarification; they give you an inquiring look “… so what is it exactly that you do?”
The title reminds me of “food supplement” industry, unlike drugs it is unregulated and anyone can claim anything from magical weight loss pill to cure for emphysema powder. In our industry too, anyone can claim the title and define it as he s/he sees fit. The title also conjures up images of a two-class software society: a lower class of software engineers who actually code and a ruling, elite class of architects that do not code or even read and understand code– and inevitably overtime lose touch with reality. Out of concern for the latter, some companies even eliminate the title altogether.
But what do exactly architects do?
I try to answer this question by answering the closely related question: What is Software Architecture?
Let me get to the bottom line first and explain later: I view architecture as a discipline that enables system to deal with change in optimally.
There are two key terms in this definition, change and “optimally”.
Let’s start with change.
Non-trivial software is a system, and like any other system is subject to change over time. The change happens in all aspects and along all dimensions: scale, cost, expectations, visibility, criticality, operating environment, competitive landscape, technology, economy, people, organization etc.
One prominent and widely known example of “change” is rapid growth in demand from a software system or what is commonly known as “scale”. Our general expectation of well architected system is to handle the scale gracefully and with no disruption. However some systems crash and need to be fully re-designed for the new level of demand (remember early days of Tweeter ?)
To be presicous, this type of scale is scalability to handle homoginous user i.e. all users of eBay were assumed to be individual buyers and sellers (no medium to large sellers), all facebook users were assumed to be college student or all users of an enterprise system could be assumed to be internal users.
Another example of change is scale of operation or non-homoginous user demand i.e. the system dose not necessary experience a change in volume of demand from one type of users, but has to handle demand from different types of users for example eBay wanting to deal with large merchants, or facebook wanting to open its platform to everyone (not just college students).
These were only two example of change in form of scale. What differentiate a well architected system from a “system that works” is how well systems respond to those changes.
The second key term in our definition was “optimally”. Optimally means the change can be absorbed and handled by the system with no (or minimum) disruption and in other words system is designed in a way that the scope of change is predetermined and extent of it is contained: no crisis, no building the system from scratch, no revolution! They system simply evolves along a smooth path. That is the essence of architecture (at least the software kind).
The pictures below depict the graceful and disruptive response to change:

This one, shows the same change scenario but one where change is handled gracefully, presumable with a well architected system.


In the next post, I will give a few examples to make this view of what architecture is clearer.