Web Performance at Surge 2012
by October 9, 2012

Filed under: Industry Insights

TR at SurgeSurge was my favorite event last year, not only because it’s located in Baltimore (“The Charm City”), but also because it attracts a great crowd of people: hardcore systems practitioners. We were lucky to sponsor Surge this year, and once again it didn’t disappoint!

The people you meet at an event like Surge make it worthwhile, but on top of that, there are even talks too! In case you didn’t make it to Surge this year, here were a few of my favorite talks and trends:

1. TCP/SSL Optimization

The best optimization for any system is doing less work; so when Bryce Howard, Chief Architect at Avalara Inc., got up on stage and announced he could eliminate 80% of his roundtrips per web request, I was immediately pumped. What ensued was a deep dive into webserver connection optimization hitting on topics like SSL false start and certificate size starting at the packets-on-wire level. Bryce showed how AvaTax improved response time on requests by 500% — without opening new colos.  Some of this was admittedly based on his having control of a client (eg. a few of these optimizations won’t work if your client is a web browser), but there’s still some great meat in there for everyone.  If your app has a large volume of small SSL requests, I’d definitely check out the talk… when they post it!

 2. Automated Fault Detection That Works

Baron Schwartz of Percona fame closed the conference in a hail of data and science, walking through various fault detection algorithms on real data sets from enormous MySQL deployments. The take-aways? Fancier is not always better—Holt-Winters won’t solve your problem by itself. Alert on metrics that matter to your system, and only on combinations of them that reveal true problems rather than just abnormalities. But don’t take my word for it,he’s got the data.

 3. Debugging node.js in Production

Few people are doing more intensive work on improving the node.js debug/ops story than Dave Pacheco, Software Engineer at Joyent. A talk that starts with the phrase “Even in the best case, when such issues are fatal” really means serious business. If you run node, definitely keep up with his work.

4. #disasterporn

The hallmark of Surge might be the grizzled veteran’s epic firefighting story. This year got off to a rocky start with an all-too-optimistic keynote speaker, so by the time we got to lightning talks at the end of the first day, the audience was ready for blood and gore.

The lightning talks were quite fulfilling in this respect, with numerous tales of true suffering and redemption. But they also covered a range of insightful and often hilarious topics: how to save Christmas for your web business; the Root Of All Problems; or a hilarious tour through Unix history to discover the origin of the weird sludge at the bottom of the /usr/bin barrel.

In true Trace fashion, I gave a lightning talk about distributed tracing theory and research – there’s plenty of DIY resources for systems hackers. I didn’t have nearly enough time to do the topic justice (wait, a lightning talk is only 5 minutes?), so that will be the subject of a future blogpost or two.

 

The Sum of All Surge

It was a blast meeting people, attending presentations and sponsoring the lightning talks. If you didn’t get one of our t-shirts or bottle openers, you’ll have to wait for next year.