HTTP and Scalable Software Systems

If you think about the World Wide Web, it's easy to imagine it as a single software system. Once you do, you realize it's the largest software system the world has ever created — probably by hundreds of orders of magnitude. It contains trillions of lines of code, hundreds of millions of servers, and billions of clients, running thousands of different programming languages. Still, it works more or less as we expect it to work. So, what made it possible for humans to create such an enormous software system? The answer is simple: HTTP!

The HTTP protocol allows us to create perfect encapsulation. The client and the server don't need to know anything about each other, besides the URL, HTTP verb, and what parameters to pass in and expect as output. This allows billions of clients to interact with each other, without knowing (almost) anything about each other. If you reduce it down to its basic components, it becomes painfully obvious that the following is a recipe for winning.

1. Magic strings (URLs).

2. Generic and untyped payloads (JSON).

3. A DNS transforming URLs to a physical address of some client/server.

You may also like: The HTTP Series (Part 1): Overview of the Basic Concepts.

They Lied to You!

The funny thing is that this contradicts some 60 years of software development theory, with strong typing, rigid classes, OOP, etc., where everything should be known at compile time. If anything, the Web's success is based upon completely ignoring every single "best practice" we as software developers have taught ourselves over the last 60+ years.

The paradox is that the above recipe, can also easily be implemented internally within our own systems. Just throw away OOP, forget all about static and strong typing, and invoke your methods as "magic strings." You'll never again experience complexity problems, with entangled dependencies, making it impossible to create a software system above some threshold of complexity, without reducing it to a big ball of mud. Imagine the following pseudo-code:

xxxxxxxxxx

/*

 * This is a method invocation, based upon a "magic string".

 * The parameters to our invocation is basically a graph object

 * Think "JSON" here ...

 */

"foo.bar"({someUntypedGraphObjectGoesHere})

All of a sudden, there are no dependencies between the place in your code where you are invoking a function and the place where you have implemented the function. There are no strong types transferred between the caller and the method, and nothing is shared, except a mutual agreement of what data to provide and return. Your dependency graph has been effectively reduced to ZERO! No OO, no problems!

All of a sudden, an in-process method invocation has been completely decoupled from the underlying method, and you have all the scalability features internally within your process, as you have with the HTTP standard. This results in software systems with the same amount of complexity as the Web, without experiencing scalability problems that normally occur long before you reach this amount of complexity.

Isn't This SOA?

No. Service Oriented Architecture is based upon having multiple servers, and/or processes. This is completely in-process. There's no need to fiddle with socket connections, server configurations, or anything that creates added complexity. In fact the "DNS" of the above "Magic string method invocations" is a simple dictionary, resembling the following:

xxxxxxxxxx

Dictionary<string, Type> _dns;

Then, you look up a type from your "DNS," like the following, and instantiate an instance of an interface, making it possible to invoke your loosely coupled function.

xxxxxxxxxx

var type = _dns["magic-string"];

var instance = services.GetService(type) as IMagicStringType;

instance.InvokeLooselyCoupledMethod(/* ...arguments... */);

At this point, all you need is a Node class, capable of passing around arguments — basically the equivalent of JSON for C#, something resembling the following:

xxxxxxxxxx

class Node

{

  public string Name { get; set; }

  public object Value { get; set; }

  IEnumerable<Node> Children { get; set; }

}

Your dictionary becomes your "DNS," the magic string becomes your "URL," and the Node class becomes your "JSON." This allows for completely independent modules to interact with each other in the same way that the HTTP standard allows completely independent servers to interact with each other, completely eliminating every single dependency between your two components in the process. And as to the speed of this? Well, it's a simple dictionary lookup and an IoC dependency injection invocation. It's more or less the same speed as anything you're already doing in your .NET applications.

Further Reading

• A Systematic Approach to Write Better Code With OOP Concepts.
• The HTTP Series (Part 2): Architectural Aspects.
• A Brief History of TLS.