To Share or Not To Share
Tuesday September 18, 2007
Hamerschlag Hall D-210
This is a practice talk for VLDB 2007.
Carnegie Mellon University
Intuitively, aggressive work sharing among concurrent queries in a
database system should always improve performance by eliminating
redundant computation or data accesses. We show that, contrary to common
intuition, this is not always the case in practice, especially in the
highly parallel world of chip multi-processors. As the number of cores
in the system increases, a trade-off appears between exploiting work
sharing opportunities and the available parallelism. To understand the
trade-off, we develop an analytical approach that predicts the effect of
work sharing in multi-core systems. Database engines can use the model
to determine whether work sharing is beneficial and apply it only when
appropriate, resulting in an average performance improvement of 20% and
2.5x, respectively, over never- and always-share policies.
Ryan Johnson is a PhD student working with the StagedDB project under
Natassa Ailamaki. He received a M.S. ECE from Carnegie Mellon University
and a B.S. ECE from Brigham Young University. His primary interests are
architecture-aware computing and high performance database systems.