3 ms·
Moore's Law still applies today; the number of transistors continues to double every couple of years. However, the transistors no longer go towards making cloc
by JoshTriplett 15y ago
Moore's Law still applies today; the number of transistors continues to double every couple of years. However, the transistors no longer go towards making clock speeds faster; instead, we get more parallel cores, and more impressive instruction sets.
And I agree that we've taken too little time to look at parallelism. In particular, our current approaches to concurrency (such as mutual exclusion or transactions) don't actually provide much in the way of useful parallelism; they just serialize bits of the code that access shared data. We don't actually have many good algorithms that allow concurrent access to the same data, just concurrent access to different data.
Currently writing my dissertation on exactly this topic: a generalized construction of scalable concurrent data structures, which allows concurrent access to the same data from multiple threads without requiring expensive synchronization instructions.
- JulianMorrison 15y agoClojure's data structures are rather good for "concurrent access to the same data". References change value, but values don't change. The references are protected by read/write locks but the values in them need no protection once they have been extracted and while they are being used.