Kevin's Weblog

On Wednesday, the pace picked up a lot. I’m going to have to condense my notes considerably to keep this post to a reasonable length.

I was pretty impressed with the keynotes this morning. Tim O’Reilly had some interesting thoughts to share on the matter of open software vs. open services. We have to think about what openness means when even if you had, for example, Google’s source code, you still couldn’t run the application (since you probably don’t have thousands of CPUs laying around). The Intel guys talked about Threading Building Blocks again. Sort of amusing as they had the “marketing guy in a suit” start the talk, and then get interrupted by the geek in a teeshirt to announce the open-sourcing. Then, keeping with the theme of concurrency, Simon Peyton-Jones talked about software transactional memory, which is basically using a declarative syntax for atomic code sequences.

Andy Lester gave a good talk on “Managing Technical Debt”. First step, you have to identify your debt: ugly code, fragile code, failing tests, TODO and XXX comments, lack of bug tracking or version control, jerks on the team or in management, lack of coding standards, etc. That’s the easy part. Then you have to figure out what the technical debt is costing you (in terms management understands), as well as what it will cost to get rid of the debt. Then start paying down your debt starting with the most profitable. (Not the easiest or most fun.) Make a commitment to not taking on new debt. If you do, you must record it somewhere and track your net “cash flow”. Stay vigilant, watch the corners, automate stuff (testing and standards). Remember: debt management is an investment.

A talk on Seaside and Rail was mostly about Seaside, and was very similar to every other Smalltalk presentation I’ve been to. Basically, it was about how the experience of Smalltalk (and the development environment) is great, and that’s all you need to know. The only real content about Seaside is that it’s continuation-based, but he didn’t address the obvious questions like: what about bookmarking, or sending links to people? If continuations are kept live forever, what about memory leaks? How do you scale horizontally?

Next I want to another talk by Simon Peyton-Jones about “Nested Data Parallelism”. The nested part is having data-parallel algorithms calling other data-parallel algorithms. Trying to do this by hand is almost hopeless. So, instead he says you should let the runtime do it for you. This is in the GHC CVS, although not released. You can write code that’s almost identical to normal array syntax but that gets executed in parallel if you have multiple processors. For example:

vecMul :: [:Float:] -> [:Float:] -> Float
vecMul v1 v2 = sumP [: f1*f2 | f1 <- v1 | f2 <- v2 :]

In this case, sumP also does the sum in parallel, (making it nested). In order to get good performance for this, you need to fuse the two loops to multiple and add in one step. GHC can also do this automatically. The basic summary of this talk is: GHC is nearly magical, and writing data-parallel algorithms is getting much easier

Let’s see... then I went to a talk about the Atom Publishing Protocol, which was a little dull.

Finally, I went to “Myths of Innovation” which was quite good and entertaining. By the author of the book with the same name, this talk was about how most of what people understand about innovation is wrong. For example, we tend to obsess about the moment of breakthrough, ignoring all the context and work leading up to it. We also ignore mistakes and failures. Another big fallacy is that people like new ideas. On the contrary, new ideas that threaten people’s jobs or established habits make them scared. Overall, this was a good talk and I imagine that I’ll pick up the book at some point.

I didn’t go to any of the evening activities. See the next post for why.

Oh, yeah, I should explain the title of this post. I think every talk this year has had at least one lolcatz slide. My feedback to presenters: one lolcatz may already be too many; having every other slide be a lolcatz is not cool.