Comment by jasonkester | Hacker News Reader

Comment by jasonkester | Hacker News Reader

We're a Microsoft shop, so the sweet spot for us is a small number of fast servers. We pay for tools and server licenses, but surprisingly less than you might expect.

Hardware: Half a cage at a colo with one fast box for web/db server, and a few cheap pizza boxes for build server and misc. This used to be two boxes in a garage behind a business DSL until Twiddla went big.

Content storage & delivery: S3/Cloudfront

Heavy lifting: EC2. We spin up a half dozen boxes each night for a few hours of processing for S3stat.

Email: Google Apps (they're terrible, but free and functional)

Tools: MSDN + EmpowerISV = $400 for all the dev tools for the shop. Small Business Server for the boxes in the cage. ReSharper, CodeSmith & RedGate's SQL stuff are actually the biggest per dev cost, but they make life so much better that they're worth it.

All in, it works out to about $500/month to keep up and running. Total, across all our sites (and there are quite a few). The colo is expensive, but it's so nice. Even with the cloud stuff where it is today, I'd still put a box in a datacenter if I had it to do over again.

"until Twiddla went big"

How big to have to change?

"half dozen boxes ... for S3stat"

What kind of boxes do you use and how much do you need to process?

(I use 2 of your services and I love it)

We moved into the Colo when Twiddla got accepted into the SXSW Interactive Awards. Good thing too, because we won our category, got written up by every major tech blog simultaneously, and spent a few days dealing with 1000 pageviews per second. This was before EC2 came out with Windows instances, so the machine was pretty much on its own. Today if the same thing happened, we could spin up a dozen EC2 boxes to deal with the spike.

That one box handled the load remarkably well, especially considering that I hadn't done much backend optimization at all, and that we were still persisting every single message to the database. It went a long way towards solidifying my faith in the MS stack, and making me scratch my head whenever I see somebody's blog fall over from a minor HackerNewsing.

As to S3stat, it runs on EC2 m1.small instances. Towards the end of the month, when each bucket takes longer to process, we'll step up the number of workers we fire up each night, and occasionally pull in a few c1.medium instances to speed things along.

It's a fun architecture, and I've been meaning to write it up for a while now. It uses six distinct AWS offerings to do its thing.

Reply on news.ycombinator.com