I understand this approach just splits shared sbitmap into per-numa-nodeSpreading the memory out does probably make sense, but we need to retainI wanted to keep the code changes minimal for an initial RFC to test the
the fast normal case. Making sbitmap support both, selected at init
time, would be far more likely to be acceptable imho.
water.
My original approach did not introduce the extra load for normal path and
had some init time selection for a normal word map vs numa word map, but the
code grew and became somewhat unmanageable. I'll revisit it to see how to
improve that.
part, but what if all IOs are just from CPUs in one same numa node? Doesn't
this way cause tag starvation and waste?