Hacker News Reader

I don't understand why parallelization of a kernel task is a good idea. Zeroing a page shouldn't take that long. Especially on modern processors. Also what else is a heavyweight task besides zeroing a page?

View on HN

The processor may be fast but the DRAM has not kept pace. So it does take a long time to zero pages. The CPU is mostly idle waiting on its write queue - but from the OS perspective is not available for other work just as if it were factoring primes.

I always thought x86 could benefit from block memmove/memset hardware external from the CPU.

Reply on news.ycombinator.com