Monday, July 29, 2013

Notes from OSCon

I spent last week at OSCon, the O'Reilly Open Source conference. It was my first time in Portland, and I prepared by watching segments from Portlandia on YouTube. I fell in love with the city, but had mixed feelings about the conference.

OSCon is big. Four and a half days of talks, with up to 18 different choices for each time slot. Topics range from Perl (it started as the Perl Conference) to “Geek Lifestyle.” My current focus on alternatives to Java drove many of my choices, but I tried to sample a few topics that were new to me. Here, then, are my top take-aways from the conference:

Invest in disk drive manufacturers

This point was driven home by Jay Parikh's keynote speech. He is the VP of infrastructure engineering at Facebook, which receives 7 petabytes of new user photos each month. Users have the expectation that those photos will always be available and never go away. In response, Facebook designed a dedicated storage rack, and a new building to house them: it will have three server bays, each holding 1 exabyte of storage.

One exabyte is 1,000,000 terabytes. Which, given 4TB drives, doesn't translate to that many physical units. But Facbook is only one part of the story. Worldwide data storage is counted in zettabytes (1,000 exabytes), and is growing exponentially. At least for the foreseeable future, traditional “spinning rust” drives are the most cost-efficient way to store that data. While there may not be a lot of profit in hard drives, volume is important. And the win will go to manufacturers that reduce electrical consumption: to support 1 exabyte of storage, Facebook is using 1.5 megawatts of electricity; a 20% improvement would be huge.

As an aside, this also tells me that Facebook is a screaming short: you can't sustain exponentially-growing costs indefinitely.

Java-the-Language really is fading

The Java and JVM track had exactly one presentation that was primary about Java-the-language, versus three that were primarily about Scala. More telling is to compare the Java track with JavaScript and Python: the latter were filled with talks about using the languages to solve problems, while the former featured frameworks.

OK, part of this may be selection bias specific to the conference. But it's another data point for my career planning.

“Clean Code” still isn't mainstream

In one talk that I attended, the presenter demonstrated how a relatively short Erlang function could be improved by breaking it into smaller pieces. While I enjoyed the presentation, I was compelled to point out that it was simply a restating of modular decomposition, a practice that has been part of software engineering since the 1970s. Or, applied after the fact, refactoring, which has been popular since the late 1990s.

I wonder if clean code is a habit that only comes from experience?

I prefer smaller conferences

While OSCon was big, with lots of sessions on many different tracks, I found it lacking: few of the talks provided insight, and many were downright boring (to the point that I walked out of a couple). This may have been due to my selection of talks, but I got a similar impression from other people that I talked to.

I think that part of the reason for this is that a conference the size of OSCon tries to cater to many needs, from the Perl programmers that were its core constituency, to people writing mobile apps. By comparison, a smaller conference (for example, ETE, which is produced by my employer) has the ability to focus: a few topics, but deep presentations on them.

No comments: