Analytical considerations for transactional cache protocols

Since the early nineties transactional cache protocols have been
intensively studied in the context of client-server database
systems.
Research has developed a variety of
protocols and compared different aspects of their quality using
simulation systems and applying semi-standardized benchmarks.
Unfortunately none of the related publications substantiated
their experimental findings by thorough analytical considerations.
We try to close this gap at least partially by presenting
comprensive and highly accurate analytical formulas for quality
aspects of two important transactional cache protocols.

We consider the non-adaptive variants of the "Callback
Read Protocol" (CBR) and the "Optimistic Concurrency Control
Protocol"
(OCC).
The paper studies their cache filling size and the number of
messages they
produce
for the so-called UNIFORM workload. In many cases
the cache filling size may considerably differ from a given
maximum cache size - a phenomenon which has been overlooked by
former
publications.
Moreover for OCC, we also give a highly accurate formula which
forecasts
the transaction abortion rate. All formulas are
... mehrcompared against corresponding simulation results in order to
validate
their correctness.