"Henry Böhlert" wrote > we're working on an application that does a lot of select > count(*) ... and thus suffers performance on SAP DB compared > to o*ther dbms. > > Searching for information about b*-trees in the internet I > found a suggestion by Simon Tatham for count maintenance: > http://www.chiark.greenend.org.uk/~sgtatham/algorithms/cbtree.html > > I also found a thread in the suse-adabas list regarding > "[suse-adabas] ADABAS Performance", where Joerg Bruehe > pointed out that maintaining these counts up to the root node > has negative impact on parallel transaction handling. > > So there are (at least) two positions: > - prefer maintaining count for optimzing query execution and > handling bags > - prefer table scan to keep update collisions low > > What is your opinion?
we agree with Jörg Brühe and neither do nor will do propagate each single row resp. its count up to the root. Elke SAP Labs Berlin
