| |||||||||||||
Priority-Boosting RCU Read-Side Critical SectionsPriority-Boosting RCU Read-Side Critical SectionsPosted Feb 8, 2007 18:05 UTC (Thu) by nix (subscriber, #2304)Parent article: Priority-Boosting RCU Read-Side Critical Sections
It's a little worrying that all of these approaches add a new task waking up 100 times a second or thereabouts. Even ignoring the extra icache hit this imposes, doesn't this somewhat torpedo all the effort that's been going on to increase the amount of time the system can go without being woken up?
(At the very least this whole thing ought to be turned off on power-sensitive devices: but in the end, won't that be pretty much *all* devices, especially embedded ones, which is one domain where realtime stuff becomes important?)
(Log in to post comments)
Priority-Boosting RCU Read-Side Critical Sections Posted Feb 9, 2007 14:44 UTC (Fri) by PaulMcKenney (subscriber, #9624) [Link] Indeed, and this turns out to be a problem for dynticks -- the CPU running the RCU priority booster will never be able to shut off its scheduling-clock interrupt. Therefore, the next step is to avoid waking the RCU priority-booster task when there are no sleeping RCU readers. In the meantime, Ingo Molnar has wisely turned the wakeup rate down from 100 times per second to one time per second.
|
|||||||||||||