TechCrunch cloud computing roundtable: Scaling apps on Facebook

>> We're being a little overly optimistic in saying that you can just take an application and put it into the cloud and this magic happens right. The reality is for the vertical applications there is a huge amount of work that goes into delivering on the linear scalability, the self-healing nature when servers and, you know, disks fail, troubleshooting, continuous change, I mean these are really hard problems. You know if you're looking at a start-up and you can figure out how to develop a software platform that makes it easier to go off and, you know, take an application and make it actually function like a real scalable cloud system that's huge because we have to spend way more energy on that then we do on the actual logic that drives the user experience.

>> Even within an application you can have very different usage and access patterns that really drive very different changes in your core architecture. I'll give you a specific example. There's about a billion status updates a month on Facebook status updates are, you know a couple hundred characters, couple hundred bytes, there's also about a billion photos uploaded a month, you know those photos are a couple megs each. The storage and access patterns of those are very, very different and we'd be silly, foolish to put them on the same backend we have completely engineered backends for each all the way down to the hardware level and the specific configuration of those machines that's optimized for cost and performance for those two different things that, you know, at the service to the consumer don't look very different. So it's not like we have two totally different demands of applications just two different usage patterns which really drive a very different architecture. You know abstraction is the great unsolved problem in computer science you know, Corbo assumed spelling wanted to waive away the network in the 90's and we all saw how that worked.

>> I do think it's very important that the apps that are being developed on your platform actually immediately meet the same scaling capabilities. So if I look at the number of, you know, very popular Facebook apps take Crayfish they built, you know, inaudible games in Flash specifically for the Facebook platform. These are such popular games that if you really need to have all the expertise to, you know, do all the infrastructure for that you would never be able to get those games off the ground. And at the same time, you know, some of those games become successful and others don't. And so as a business wanting to minimize the risk in terms of what you have to do to build out your infrastructure, you know, you're much better off with going with infrastructure as a service for that case because otherwise those apps would never get off the ground.

Sound effect

==== Transcribed by Automatic Sync Techologies ====