Thinking of the backend of websites like Amazon, Google, Facebook, etc, where they have to serve hundreds of millions of users, have great response time, security, and uptime, I was wondering how many developers exist that can design and execute such a system? How common or rare is this capability?<p>I thought a quick poll here on HN might be a good indicator.<p>So, are you capable of designing and executing a backend that is on par with the ones used by the major players mentioned above?
The thing to remember here is that these systems, to the point they are today, were built by teams of people over lengthy periods of time. There are many pieces that have to be looked at from the wiring infrastructure, location, network equipment, monitoring solution, server solution, backup solution, security, OS, etc... While an individual may be capable of putting together an comparable infrastructure with only one or a few servers, once you cross a threshold with requirements and capabilities it would be almost unbelievable to find a single person who could complete the project to the same scale.
While, the systems I designed and built may not be considered Google or Amazon scale, I have build some of the largest travel systems the web uses. I have build systems that not only the biggest use with hundreds of thousands of request per second (hotels.com, travelocity, Expedia, Orbits) but thousands of other smaller clients where supported simultaneously. Anyways to get to the point there where two major decision that we made early on that saved us years and millions of dollars in frustration. Those two where statelessness and decoupling.<p>Statelessness is important because it drastically simplifies your software architecture and ejects the concern of scalability into the infrastructure. This is a good thing because with a good automated install strategy new capacity can be online in minutes (now with the cloud seconds). When you are stateless you don't have to worry about inter-application communication among nodes, rather you can have dead simple load strategies like load based, round robin, load balancing where the most under utilized node gets the next request. Statelessness is paramount to scalability, at least to cost effective scalability.<p>Now on to decoupling, I don't get too caught up in the enterprise buzz word bingo, but there is one that developers should pay attention to and that is SOA. SOA is just a name for a good architectural strategy that has been employed for years at the big guys. Summed up SOA is just a form of decoupling. If a system does an independent task then that system should be decoupled from the system as a whole. You should not think hey lets build a reservation application you should think lets build an availability service, a booking service, a concierge service. And these services as a whole will be assembled together by a UI into a work-flow that allows people to make online reservations. REST is just an extension of this in a well defined and intuitive manner.<p>This allows the services to be used in new and novel ways, not only that but it isolates the system and makes capacity planning very easy. If you see a good deal of load on your availability service or your search service you can scale that area with precision, allowing you to put the horse power where it counts. As well with a decoupled system, that has well defined interfaces if a portion of the system in unreliable the entire piece can be snapped out and replaced with no effect to the larger system as a whole. It becomes very easy to trim out the rot in a decoupled system.<p>Going further, and on a side tangent, a decoupled system is easy to bring in contract development capacity in to work on it. You define the interface and what the component should do, and you can hand those component documentation off to a third party to build with far less risk than "hey build this whole system". This becomes critical if you experience the two year growth rate that we did.<p>I would say those two choices above all other allowed our small start-up the ability to scale on demand and grow from a handful of niche client to supporting every major travel system in a two year time frame.<p>Had we not made those choices we would certainly had a few year time-frame that would have been painful while we redesigned the architecture to handle the scale at which the business grew.