sched_yield() makes OpenLDAP slow
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): 20050818010703.GA13127@nineveh.rivenstone.net.suse.lists.linux.kernel
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
Bernardo Innocenti <[email protected]> writes:
It's really more a feature than a bug that it breaks so easily because they should be really using futexes instead, which have much better behaviour than any sched_yield ever could (they will directly wake up another process waiting for the lock and avoid the thundering herd for contended locks)
-Andi
Quoted 7 lines · Andi Kleen
Bernardo Innocenti <[email protected]> writes:
It's really more a feature than a bug that it breaks so easily because they should be really using futexes instead, which have much better behaviour than any
sched_yieldever could (they will directly wake up another process waiting for the lock and avoid the thundering herd for contended locks)
Actually, I believe they should be using pthread synchronization primitives instead of relying on Linux-specific functionality. Glibc already uses futexes internally, so it's almost as efficient.
I've already suggested this to the OpenLDAP people, but with my limited knowledge of slapd threading requirements, there may well be a very good reason for busy-waiting with sched_yield(). Waiting for their answer.
Signature · 2 lines
// Bernardo Innocenti - Develer S.r.l., R&D dept. \X/ http://www.develer.com/