Notes:
TODO:
(1) make the buddy and deferred lists singly-linked
(2) find a better home for /proc/fraginfo somewhere in /proc/
(3) analyze cache performance
(4) get real benchmark numbers out of the thing as opposed to the
one-shot few-rounds stuff I've done thus far. Thus far
benefits for local network bandwidth and latency have been
seen in lmbench and netperf.
(5) collect statistics on the allocation rates for various orders
(6) collect statistics on the allocation failures for various orders
(7) collect statistics on allocation failures due to frgamentation
(8) collect qualitative information on long-term fragmentation
Follow-on patch TODO item:
(9) integration with upcoming gang allocation patches
(10) gang preassembly and bulk transfer to and from per-cpu pools.
|