Paul, > -----Original Message----- > From: Paul Reeves <pree...@ibphoenix.com> > Sent: Tuesday, March 19, 2019 12:50 PM > On Tue, 19 Mar 2019 16:03:21 +0000 > "Leyne, Sean" <s...@broadviewsoftware.com> wrote: > > > > > There is very little documentation available on the Classic lock > > > > manager related settings, which is why I ask the question in this > > > > forum. Interested in v2. 5 limitations, but details/differences > > > > for other versions would be appreciated. > > > > > > I assume you mean hash table slots: > > > > > > const SLONG HASH_MAX_SLOTS = 65521; > > > > That was what I was looking for? > > > > Is there any reason why for x64 builds that number could not be > > increased? > > > > > As I understand it, a high number of slots will reduce the queue lengths at > the > cost of checking a lot of empty slots.
But if the slot is empty, then there are no other connections working on the slot, so no contention. No? A given page would 'hash' to a single slot, no? Further, when the lock manager needs to check for a lock, the fact that the slot is empty means nothing for performance, no? > I did some tests a while back which showed that hash slots greater than > around 8000 actually decreased performance. But that was with the tpc-c > benchmark. But that seems counter-intuitive... using more slots means that the locks are being spread out into a larger number of buckets -- providing for less "collisions". Firebird-Devel mailing list, web interface at https://lists.sourceforge.net/lists/listinfo/firebird-devel
