Bugzappy's Blog

With the beta release on May 17 of its Monitoring, Auto Scaling and Elastic Load Balancing services, AWS becomes what is was meant to be from the very beginning.

Without these services, AWS is, when you think about it, only marginally different from any other Virtual Private Server (VPS) hosting service (slicehost for instance).

Today AWS is truly breaking new ground: every major aspect of a scalable web application infrastructure is now virtualized, and to a substantial extent, automated. That is a paradigm shift that the likes of GoGrid have evidently foreseen but will have a heck of a time keeping up with, for sheer lack of resources and experience.

The challenge faced by Amazon Services, however, is to remain focused on what really matters to web application builders. SimpleDB and Simple Queue Services, for instance, have been available from nearly the beginning, but belong to the realm of specialized programmatic web services, not infrastructure. As such they had to fit very nicely into a specific application architecture in order to be of any use. To boot, typical web application architectures had simply no use for them. What a waste of time when the world was starving for load balancing.

Dedicated server, virtual-private server, and cloud hosting have seen very rapid evolution over the last 18 months or so. The industry is maturing rapidly and hosting providers are jockeying for top positions in this promising market.

If you are a serious LAMP application developer you have or you will soon run into the need to set up, configure and tune the entire ’stack’. Without going into issues of scalability, here are 8 providers that I have researched. I have only used a few of them myself, so I cannot compare their quality. However the table below might help you start your own research.

I have reduced the ’specs’ to only 2: RAM, and outbound data transfer allowed per month. I list the prices for 3 configurations of these 2 specs:

1. a “developer’s box” with 1 GB RAM, and only 100GB outbound data per month
2. a “dumb-but-busy web server” with 256 MB and 1GB outbound data
3. a “workhorse server” with 1GB RAM and 1GB outbound data.

One thing worth noting is that for many web applications, outbound bandwidth is a lot more expensive than CPU and RAM.

Do the math: you can easily imagine that a server with a decent CPU, 1GB RAM can generate 1 MB of dynamic content per second (let’s assume  24/7 for simplicity). That will cost you roughly $100/mo with the below providers.

In this 24/7 1MB/sec scenario, the total bandwidth usage over a month will be 2.5 TB (1 MB * 3600 * 24 * 30). Nowadays most providers give you 1 TB for free, and so you will have to purchase the additional 1.5 TB at the rate of $0.15 to $0.50 per extra GB. Do the math…

At minimum this will cost you an eye-popping $225 on top of your $100 or so! It’s just like with cellphones. They are very cheap considering what they can do, but actually using them can become very expensive.

If you are seriously considering scaling out an application on one of these hosting providers, you will need to choose whether to use software or hardware load balancers. The software type is easy to come by: rent an extra box, deploy your favorite load balancing software solution on it. However true performance is delivered by hardware load balancers; there is a LOT of variability in the availability, quality and pricing of hardware load balancers at the various hosting providers.. so analyze your problem and design your architecture carefully before committing to one provider.

[1] Joyent Accelerators come with Open Solaris and 10TB/mo bandwidth

[2] The Planet dedicated servers come with 1.5+ TB/mo. Also see the Bargain Bin for older, discounted server models.

Disclaimer: I compiled this information in April 2009 and I believe it to be accurate; however I make no representation whatsoever on behalf of, or about the services and products of any of the companies listed here. Please check directly with them for their latest specifications and pricing.