Friday, March 25, 2005

Behind the scenes at Findory

Sometimes the impact of our work at Findory is obvious. When we launch a new feature, such as source pages, everyone immediately can see what we've done.

But sometimes our work is more subtle. Most of our work this month was on the backend systems. Sure, you might notice that the site is a little faster, but the effect is subtle, especially since we already served our personalized pages in under 100 ms. If you were very observant, you might notice that we partitioned off our RSS traffic to new servers, but since it all just works like it's supposed to, you probably wouldn't notice. And that's okay. We like it when it all just works.

But work on the backend systems can be satisfying. We substantially improved the quality of the unpersonalized Findory front page recently. Sure, Findory is about personalized news, but we want newbies to Findory to immediately see interesting and useful content, and our latest improvements will make Findory more attractive to new users. It's satisfying when we make a change that we think helps our readers find the news they need.

We completely rewrote our deployment system. When you have a cluster of servers, you want to be able to image the servers and put new content on them quickly and easily. This allows you to scale easily and rapidly as new traffic floods in. Our old system was a little too manual and unreliable for our taste, so we made it nice and bulletproof.

Ahh, yes, deployment systems. Both Alex and I have done this before. Way back in 1997, ran on one big piece of iron, a multiprocessor DEC Alpha box. It might be hard to believe now when clusters are seen as the obvious solution, but, in 1997, web server clusters weren't all that common. I worked on a team of two to split Amazon's website across a cluster of four boxes, a job that mostly involved ferreting out assumptions that there was only one box, but that also included writing the deployment system to push and pull content from the boxes. It was the first step toward Amazon's current cluster architecture, and it was exciting to be on the bleeding edge of it. More recently, Alex wrote a deployment system for some of Amazon's famous web services. We know scalable systems, and it's fun to be building them again at Findory.

Finally, we've been fielding a lot of interest in Findory from the search giants, newspapers, and VC firms. Perhaps some of this is that the exponential growth we've been seeing is now visible in less accurate sources like Alexa. Perhaps some of this is the growing realization that the future of news is personalization. Either way, we're excited to be part of that future.

No comments: