On 10/29/2013 02:25 PM, Steve Smith wrote:
I'm wondering why VM compression hasn't caught on sooner? Has the
physical memory curve kept ahead of it well enough, or is the
compute-cost simply too high?
Most people now have cores to waste, so why not have them do something
useful? The reason people have cores to waste is that on-chip memory
is limited (some new Haswell & Power chips have eDRAM for graphics,
aimed to compete with GDDR5, but this is all well below a gigabyte in
size), and other than that there isn't the bandwidth available per core
that would be optimal. This is because the fabrication is mostly 2D,
and wires go out the bottom on a limited set of pins/routes. Hopefully
soon pins will go out the top too (e.g. stacked memory).
I think there is a scale free aspect to this. RAM vs. swap is addressed
today. L3 to eDRAM or eDRAM to RAM might be another bottleneck that
could be addressed by this. So it's worth having in the operating system.
Marcus
============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
to unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com
