On Fri, Dec 14, 2012 at 2:29 AM, Robert Haas <robertmh...@gmail.com> wrote:
> On Thu, Dec 13, 2012 at 3:03 PM, Andres Freund <and...@2ndquadrant.com> > wrote: > > It moves a computation of the sort of: > > > > result -= vacuum_defer_cleanup_age; > > if (!TransactionIdIsNormal(result)) > > result = FirstNormalTransactionId; > > > > inside ProcArrayLock. But I can't really imagine that to be relevant... > > I can. Go look at some of the 9.2 optimizations around > GetSnapshotData(). Those made a BIG difference under heavy > concurrency and they were definitely micro-optimization. For example, > the introduction of NormalTransactionIdPrecedes() was shockingly > effective. > The two commits coming to my mind are: - ed0b409 (Separate PGPROC into PGPROC and PGXACT) - 0d76b60 (introduction of NormalTransactionIdPrecedes) Those ones really improved concurrency performance. -- Michael Paquier http://michael.otacoo.com