
330 Big Data Computing
Tests with Locality of Disk Access
A crucial factor for the performance of HYBRIDJOIN is the distribution of
master data foreign keys in the stream. If the distribution is uniform, then
HYBRIDJOIN may perform worse than MESHJOIN, but by a constant fac-
tor, in line with the theoretical analysis. Note, however, that HYBRIDJOIN
still has the advantage of being efcient for intermittent streams, while the
original MESHJOIN would pause in intermittent streams, and leave tuples
unprocessed for an open-ended period.
It is also obvious that HYBRIDJOIN has advantages if R contains unused
data, for example, if there are old produ