From: NORLMN::EVERHART 3-APR-1997 08:06:09.60 To: STAR::JORDAN CC: EVERHART Subj: RE: Some good feedback from Glenn. Karen, Paul, care to reply? -steve Greg - (It's Glenn (2 N's)...) Thanks for your letter; it clears a lot up. One bit that occurs to me is that the notion of having the master tree act as a bunch of trees instead, each coordinated by a different spinlock, could be worth keeping as a thought. Given a name I'd think there might be a way to tell which such was being used. Might make for a number of hash tables based on some top level hash that tells which to use, as a for instance, and the count could be a parameter that ultimately might want to be settable locally as well as globally... in case local lock access in SMP gets to need finer grain spinlocking. I hadn't thought much about other galactic lock situations but I wonder if some replication via a priori partitioning of locking might be a useful tool in those too. I'll need to reread the (goddam long!) IR again. This stuff didn't start occurring to me till I got to locks, way near the end (by which time I was getting rather sleepy...) Glenn Everhart 4/3/1997