A partition-aware application is an application that can
make progress in multiple connected components. In this talk, we examine a particular
partition-aware application to evaluate properties provided by different partitionable
group membership protocols. The application we examine is a simple resource allocation
problem that we call the Bancomat problem. We define a metric specific to
this application, which we call the cushion, that captures the effects of the
uncertainty of the global state caused from partitioning. We solve the Bancomat
problem using three different approaches for building partition-aware applications. We
compare the different group membership protocols in terms of the cushions they provide
when solving the Bancomat problem.
This talk covers work that was done in conjunction with
Jeremy Sussman.
|