Managing the amount of visitors...

Question

Here at SO there's a lot of focus on designing web applications but I seem to be missing something in all these discussions. It seems as if everyone is focusing the design of their site to allow lots and lots of visitors to visit it, thus putting a huge strain on the server. But what if someone is designing a site just for friends and family? Or just an in-house site that's just for the 500 employees of a company? While it's all web design, there's still a huge difference in designing sites like Google Search or StackOverflow and sites like www.geocities.com\alex\myfamilyhistory.html or something similar. (Wow, I remember Geocities. I must be old!)

So, my question is simple: when designing a new website, where would be the most important differences between managing a site for 50 visitors, 5000 visitors and 500.000 visitors per day?

(Well, hardware-wise, just add more servers. But I'm focusing here on software design only!)

Answer 1

Software-wise, you should always think your design and architecture through not to limit yourself with any abstract visitors limit.

Use commonly-known best-practices. Avoid costly string operations. Create a decent database schema. Put indexes and clustered indexes where appropriate. Avoid quick-hacks based on the consolation that nobody will notice because you don't have that many visitors yet.

Then when the time comes, you will discover you only need to get more hardware instead of completely rewriting your application.

Answer 2

There are many design patterns that will never be used in applications with small number of users or requests, most of these patterns are patterns related to load distribution.

For example, one of the design patterns for load distribution is having your data in clustered (tree structure) servers. Every request is directed to one server, which in turn says that the data of this user is in a given cluster. it doesn't have the data itself, it just redirects the users to second cluster level.

The servers in the second cluster level (or first tree level) could have the data (or part of it) or could redirect the user to a third cluster level. the same thing goes on an on until the request reaches a server that already has the data of the user, so the response is generated and sent back to him.

This design pattern (sometimes called routing) is an example of a scenario that you will never use unless you have an application with massive number of requests. For example gmail uses this design, and the last failure of gmail's web interface was because the routing servers got overloaded with requests.