Old but good bit from Brewer demonstrated a good thread implementation is able to handle as much concurrency as a good evented program. Bonus: Russ Cox, now working on Go, is acknowledged for his assistance.

Intel's C/C++ extensions that allow compilers to introduce parallelism in an automatic and safe way. These are syntax, not libraries. The hyperobjects remind me of something between software transactional memory and Dynamo-style conflict resolution.

A superset of the TBB, these tools (even without using ICC, apparently) help exploit data-level parallelism.

Goals for the Intel Cilk GCC branch.

I'm guilty of conflating the terms "concurrency" and "parallelism." This article is more clear than most about the precise difference.

An explanation of a divide-and-conquer algorithm in the context of a very high-performance parallel I/O system. It processes 235 MB/s even as the disk can only deliver 150 MB/s. This validates my experience that global buffers are pretty much guaranteed in high-performance systems.

Back at it, this time with a beast of a server, a dataset that won't fit in memory (joy of joys!) and (inexplicably) more OCaml.