Eric.Saxe wrote: > Li, Aubrey wrote: >> This is also what I was thinking. But when the event driver set the >> policy to a per CPU basis, this property may not be bringed into >> cmt_balance(), because cpu_choose() will choose cpu again. The cpu >> will be different. So I think a per thead basis is better than a per >> CPU basis. >> > cpu_choose() will likely choose the same CPU again, but then the > choice may be overridden by the call to cmt_balance() a bit later in > setbackdq()... >
If a thread has been waiting longer than the *cpu_rechoose* value(3 ticks), the potentially benefiting from a warm hardware cache is diminished, so the dispatcher will choose cpu again. Back to the original issue, do we need a per-thread basis cmt policy? Thanks, -Aubrey
