TaskSpaces: A Software Framework for Parallel Bioinformatics on Computational Grids - Parallel Computing for Bioinformatics and Computational Biology - page 656

concept allows tasks to be decoupled both in space and time. The distributed comput-
ing process is decoupled in space, as the application, task and results bags, and the
various worker processes may reside on a heterogeneous collection of machines that
are connected by a network but that are otherwise widely geographically distributed.
This decoupling allows flexible topology for the computation, permitting automatic
configuration based on the availability of worker processes. The distributed comput-
ing process is also decoupled in time: as spaces are persistent, tuples are persistent
while resident in the space, and processes can access tuples long after the depositing
process has completed execution.
Figure 26.3 shows a conceptual deployment diagramof theTaskSpaces framework.
TaskSpaces uses an event-driven model. On startup, worker processes register with a
task bag. The application process sends subtask objects to the task bag, and the task bag
sends those task objects to available workers. The task bag acts as a “superqueue,”
and thus alleviates the problem of scheduling when multiple supercomputers with
different unsynchronized queuing systems are used. Scalability is inherent because
users may put several applications in the task bag at the same time and the grid
operator can add “worker farms” when needed. After a task is processed, the worker
puts a result object in the result bag, from which the result objects are collected
for final assembly by the application. TaskSpaces is implemented in Java, providing
a standard, platform-independent interface to the grid system and exploiting Java's
built-in networking and security features.
Figure 26.3 TaskSpaces framework deployment diagram.