Surge 2012 Postmortem

2012-09-28 18:35:11 by jdixon

The curtain has lowered on another couple days of scalability lessons and "disaster porn" at this year's Surge conference. Despite my initial misgivings that the registration fees were too high, the conference organizers have once again put together an experience that is quite possibly the best among all technically-oriented events.

As mentioned previously, the #monitoringsucks BoF was probably best described as a "group therapy session" for monitoring addicts. Some excellent points were made that have already got me thinking about how data forecasting and search could be improved upon. I think the word addict cannot be overstated here; I had numerous attendees ask me about the possibility of a follow-up BoF to continue where we left off (as if I were some sort of Pied Piper of Monitoring).

The formal conference started off with a sputter. Although Robert Treat served as a great emcee to the opening keynote, the keynote itself was underwhelming and was the subject of wide criticism on Twitter and in casual discussions. Mike Gualtieri, an analyst at Forrester Research, made a number of questionable claims about the future of tech; in fact, his content seemed less suited for an audience full of system architects, engineers and operations trailblazers who are dictating the future of Web Operations scalability, and more tuned for a collection of Fortune 500 CEOs who are testing the waters for technology trends.

Fortunately this hiccup was short-lived and quickly rectified. Morning sessions from Pinterest and PagerDuty speakers were well-received. The Thursday lunch was very good. Afternoon sessions, in particular those by Tom Daly of Dyn and David Pacheco of Joyent, were fantastic and full of valuable insights. Closing out the Thursday sessions was Artur's always-entertaining take on CDNs. Afterwards the audience was treated to a moment of restrained brilliance as Tom Daly dissected and debunked Artur's claims against using Anycast in public networks. Unless you were there it's hard to describe, but suffice it to say you'll want to look out for this exchange when the Surge videos become available.

This year's lightning talks were exceptionally high quality and thoroughly entertaining. Scott Sanders' talk on the importance of generalists and Bryan Cantrill's trace of the history of the ta UNIX command were both priceless. The after-party was a fun event; drinks and music were provided, while attendees mingled and kept the exhibitors hopping. Afterwards a small group of us made our way over to Mick O'Sheas for pints and Cricket (darts).

Apparently I had too much fun at the bar that night, because I overslept the next morning. I woke up in time to clean up and hear the 2nd half of Bryan Cantrill and Brendan Gregg's talk on data-intensive real-time semantics (DIRT) in web applications. Every talk I've ever heard from Bryan has been entertaining, and this one was no different. This seemed to highlight a general theme that heatmaps and/or histograms are increasingly important for data analysis. Theo Schlossnagle and Brian Clapper continued this theme in their talk on diagnosing distributed systems.

After a quick step out for lunch with friends, I reconvened for a short while to give demos on Descartes and Tasseo. This was a good chance for me to gauge (excuse the pun) user response to an open-source dashboard that was designed from scratch to improve the user experience among tools in the visualization space. Everyone who experienced it seemed genuinely impressed with its capabilities and interface. This seems to justify a lot of the decisions I've made within the application; I'm excited to keep pushing this forward and see if adoption increases.

My last session of the even was Matt Graham's talk on continuous deployment. As you might expect from an Etsy engineer, this talk was full of insights on Etsy's well-known deployment strategies. Of particular interest to me was a metric known internally as "bughours" that is, the factor of hours that any collection of bugs are publicly exposed to customers. I would personally love to see this sort of metric correlated with more business-centric data such as revenue or sales per unit. Etsy's deployment model is a perfect example of decreasing the surface area of bugs and outages by iterating quickly and shipping often.

Lastly, I should mention that the internet problems with last year's event were completely rectified. The wireless connectivity worked well throughout the conference venue and without interruption. The OmniTI team deserves a lot of credit for learning from their mistakes and delivering a much-improved experience, even by their already-high standards. I'm really excited to see what they have planned for next year.

Add a comment:




max length 4000 chars