Experiments on the Single-Processor Architecture

   On the single-processor architecture, uniform lifespan partitioning surprisingly produces the best performance results in many situations as long as the value for m is high enough (i.e. approx. ). For primary underflow and minimum-overlaps partitioning the best choice of X_R and X_Q values for a join is given by the rule

for , according to the experiments of section 10.4.

Condensation is similarly successful on the single-processor machine as on the parallel architecture. Primary underflow as well as primary minimum-overlaps partitioning is hardly affected by condensation approx. for . In some cases we even observed some minor performance improvements (see figures 10.46 and 10.47). This is again very encouraging due to the many advantages that can be drawn from smaller IP-tables.

In contrast to the parallel case, the negative impact of black-out preprocessing on the join performances were only minor (around 0.6% performance decrease for the worst case). In some situations, however, improvements of up to 3.3% were observed. This is again relatively modest.