Seth Eliot's Blog

One thing I had forgotten in my time away from Amazon was that their code management, packaging, and deployment systems are magic. Seriously impressive internally built tools to manage a large, but usually decoupled codebase across thousands of servers in data centers around the world. Magic indeed… sometime black magic, but magic nonetheless.

There are pretty strict limits on what I can say about internal systems, but on the deployment piece at least the cat is out of the bag. Werner Vogels, Amazon CTO talks about Apollo, the Amazon deployment system here.

By comparison, the other massively scalable deployment system I know about is Microsoft’s Autopilo, which you can read about here,.  Autopilot is good for what it is, but Apollo has the advantage in many areas including interface (usability), configurability, packaging, and versioning, which is not surprising given its 10 year head start.

And those systems are just deployment.  My ability to search code, have code dependencies just handled for me (at development, build, and debug times), and to build code in a consistent trackable way is simply impressive at Amazon, reflecting Amazon’s long history in the software services space.

Finally my ability to request new servers at Amazon (generally EC2 virtual servers) is as easy going to a website and filling out a form.  I remember in my early days at Microsoft having to select SKUs and meet with a vendor to place the order. Then a week or so later I got a phone call telling me my servers were on a pallet on the data center loading dock…. what did I want done with them :-)

Amazon has done DevOps long before that word existed.  Like Microsoft, any company making the move into software services learns quickly how important that is.