Speaker: Srikanth Sastry, TAMU
Date: Friday, April 23 2010
Time: 1:00PM to 2:30PM
Location: 32-G631

ABSTRACT:

Ever since its introduction in the 1980s, partial synchrony has
gathered much interest and popularity as an effective mechanism to
circumvent the impossibility of solving classic problems in
distributed computing in the presence of crash faults. Among the
several algorithms designed for solving problems in the canonical
models of partial synchrony, very few have actually been implemented
in empirical systems. One of the primary reasons for it may be that
these idealized partially-synchronous models specify an extremely
`well behaved' system that favors algorithm design, but the guarantees
provided by these models do not bear sufficient fidelity to the many
empirical systems on which these problems need to be solved. This talk
introduces algorithmic techniques that can be used to bridge this gap
between idealized partial synchrony and empirical systems. The
proposed methodology involves (a) constructing `system services' on
existing empirical systems, (b) using these system services to
implement a distributed scheduler that provides the guarantees
specified in the idealized models of partial synchrony, and (c)
scheduling/executing algorithms designed for such idealized models
using the aforementioned distributed scheduler. The talk focuses on
the interesting challenges posed by such model transformations and
investigates the role of time and its measurement in specifying
partially-synchronous systems for solving time-free problems.