> >> PS Snapshot (concurrency) transaction guarantees that once read > >> record could be read again and will be the same. So, in theory, we > >> can just re-read same data page when looking for next record. > > > > Why "in theory"? > > Because i want to be careful and don't want to make too early conclusions > :) > > This (raw) idea should be considered from the all possible points of view > and, ideally, we need to find a way to make same optimisation for every kind > of transaction.
I would argue that not every transaction can have the same optimization. We may need to apply optimizations on a "common cause" basis, which would allow for some modes/transactions to use Optimization A vs Optimization B vs. no optimization. Handling a SELECT for a Snapshot transaction is completely different that an UPDATE statement for a read-committed transaction. > > It would seem that a significant performance improvement could be > yielded by changing the current functionality. > > It could be not so significant, especially with shared cache in mind :) Not needing to access the cache at all, therefore by extension no page coordination, would have to provide a benefit. Sean ------------------------------------------------------------------------------ RSA(R) Conference 2012 Save $700 by Nov 18 Register now http://p.sf.net/sfu/rsa-sfdev2dev1 Firebird-Devel mailing list, web interface at https://lists.sourceforge.net/lists/listinfo/firebird-devel
