On March 9th, Scott Baker, our Director of Systems Engineering and Operations, gave a keynote presentation at the Cloud Connect conference in Santa Clara, CA (http://www.cloudconnectevent.com) about how we think about and use cloud services. Scott offered some great insights on the cloud and what it means to a company like Eventbrite, and we asked him to elaborate for the blog.
When you use Eventbrite, you expect the service to be fast, reliable, and just there. You don’t want to think about how the systems are working underneath; it should just work. That’s how we look at our relationship with Amazon EC2 services, where Eventbrite’s server instances live. We don’t know what hardware lurks beneath our server instances; in a sense, we don’t care. We are designing our infrastructure to deal with issues that could happen below the veneer of what Amazon exposes to us.
In the old world of Web 2.0 websites, my teams and I spent lots of time tuning server and network gear configurations to work with our applications. By tweaking hardware specs and playing network games, we got increases in performance and scalability that allowed us to not worry about the software. By using EC2, we are liberated from worry about the hardware layer and can focus exclusively on making Eventbrite the application perform better.
Also, running hardware datacenters sucks a lot of time out of your life, time that could be spent analyzing how to make our software run better. At Digg, the systems and network engineering team worked very hard to get the hardware, cabling, network, power redundancy, rack layout and inventory optimized. Using a cloud service, that all goes away. I know I have no control over any of that and that we should construct our server instances in way that it doesn’t matter what Amazon buys or deploys for hardware; our stuff just works.
At Eventbrite, we’re in the process of taking our initial startup infrastructure and scaling it. To make sure that the tools we provide our users are always available, accurate and speedy, we’re in the middle of a process to create our infrastructure for the future, or at least until we decide we need something better than what we can see today. Part of that is automating the creation of our EC2 instances, be they databases, web servers, search servers, Haddoop clusters, etc. We can spin up new server instances rapidly and change configurations (hopefully) as quickly as our development teams need them.
The point of this presentation at CloudConnect was that the ideas of building scalable applications remains pretty much the same over time but that the forms of implementation change. Timesharing and cloud computing are twins, from different points of time. In the future, we’ll be sharing more of our technical exploits. Stay tuned. If you have questions, I’m available at http://www.quora.com/Scott-Baker-1.