The basic idea is to write distributed systems in a high-level logic language; essentially, to treat writing a distributed algorithm as a distributed data management problem ("data" being the state of the computation).
Very interesting, the results are already impressive (hot stand by namenode). I've previously skimmed over the Declarative Networking paper (http://berkeley.intel-research.net/dgay/pubs/06-sigmod-overl...) after seeing an article about in a recent CACM.
Are there any plans to open source any of the work (either the JVM implementation, the C implementation or some of the products, e.g. hot-standby NN)?
Thanks! We basically open-source everything we do; the problem is committing the resources to support stuff and polish it into a usable product, which unfortunately doesn't always happen.
We haven't been polishing the JOL/BoomFS work, because we're currently working on a brand-new language ("Bloom") that should be much more programmer-friendly; as part of that, I'm writing an efficient C-based dataflow engine. If you have any more questions, feel free to drop me an email.
In my case, network servers that need small-packet/high-throughput and extremely low latency (telephony stuff). Microprocessors too (AVR mostly) for various stuff.
Not yet--though I'd like to dream up a project/excuse at work to justify tooling up for it. (I've messed around with some basic FPGA and CPLD kits at home, but nothing in depth.) The gates per dollar ratio on even the low-end FPGAs is getting to be really exciting.