4 ms·
Most mainstream programming languages are designed as if we are all targeting 70s von Neumann machines. When applying these languages to e.g. highly distributed
by grumpyprole 4y ago
Most mainstream programming languages are designed as if we are all targeting 70s von Neumann machines. When applying these languages to e.g. highly distributed or concurrent architectures, they are no longer a good fit. It's great to see at least an attempt at innovation rather than just a Java clone, which is all Google and Microsoft have ever offered so far.
- astrange 4y agoI’ve never seen a program that would magically be better on a distributed architecture if it was written in a different language. Programs run on a single core because they have a single core’s worth of work to do. Autoparallelization, like autovectorization, doesn’t work.
- grumpyprole 4y ago> Autoparallelization, like autovectorization, doesn’t work. I think concurrency is more of an issue for games and was the example I gave (Haskell's transactional memory for concurrency is a great example of what pure-functional buys you). Nethertheless autoparallism can work, if you again are prepared to accept more constrained declarative languages. Apache Spark is a great example, it offers a constrained functional language with maps and folds, that is autoparallized across a cluster of machines. The research language NESL (nested data parallelism) is even more impressive.