Archive for October, 2010

Reflections from SPLASH

Tuesday, October 19th, 2010

Right now I’m attending SPLASH , the re-branded version of OOPSLA. I must say I’m amazed how ‘concurrency’ means ‘JVM parallelization’ for most of the guys around here, like there would not be any other language or run-time system around.

Nevertheless so far the discussions were really interesting. I particularly appreciated the open discussions in the well attended (full room) Concurrency for the Application Programmer workshop, where I also had a presentation (see the webpage for the full paper) . Unsurprisingly, the discussions revolved around familiar issues:

  • how to manage heterogeneous hardware?
  • design for power efficiency
  • how to hide the complexity of synchronization?
  • what are the most common concurrency bugs and how to deal with those?

Somewhat surprisingly, there was very little about automatic parallelization – it seems that the community is finally realizing that it’s a dead-end in general. It was interesting to hear the JVM gurus agree on how garbage collection really sucks and how much effort goes into fixing it, especially in a multi-core setting.

However, there was a lively discussion around DSLs and even an ad-hoc presentation by Stanford on their work was arranged – it was really nice to see this kind of open minded, flexible approach to arranging a workshop that can help you leave with something useful rather than just the memories of x paper presentations you listened to.¬†Speaking of which: Stanford will have an interesting presentation on what they call ‘language virtualization’ the coming days. They will focus on how to make a host language transparent to an embedded language so that the embedded language feels and can be used as a stand-alone language, the same way as VMs can be used as ‘real’ hardware. I’ll get back on this after the presentation, but it struck me like a fresh idea.