[Haskell-cafe] Reducing Latency - Reducing the Size of Larger Data
caseyh at istar.ca
Fri Nov 27 14:02:48 EST 2009
An idea to reduce memeory latency, is to have a complex dual/quad/etc.
core CPU surrounded by four or more simpler CPUs.
The simpler CPUs have a simpler instruction set and do less context
So any operations (e.g. folding, filtering) that reduce the size of a
larger data structure and that can be done on the simpler CPU
instruction sets are shunted off to the simpler CPUs.
So, the complex CPUs receive less data and therefore can have better
cacheing performance, less latency and better throughput.
More information about the Haskell-Cafe