JOLTS : checkpointing and coordination in grid systems

Abstract

The need for increased computational power is growing faster than our ability to produce faster computers. Already researchers are proposing systems that require peta-flop capable super computers, a far cry from what is currently capable. To meet such high computational requirements, networks of computers will be required. While it is possible to network together computers to achieve a single task, making that network more flexible to handle a multitude of different tasks is the promise of grid computing.

Grid systems are slowly appearing that are designed to run many independent tasks, and provide the ability for programs to migrate between machines before completion. However, these systems lack coordination capabilities. Many grid systems/environments allow multiple tasks to communicate/coordinate with each other based on various paradigms, but don't provide migration capabilities.

This thesis proposes a system, called JOLTS, that attempts to fill a gap by providing both checkpointing and coordination capabilities. The coordination model offered by JOLTS is based on the Objective Linda coordination language, with some additions. This thesis will show that the object space model is an effective form of coordination and communication, and can effectively be combined with checkpointing capabilities inside the same grid system.