DRMS objectives

We are designing and developing the Distributed Resource Management System (DRMS) for parallel and distributed systems. Objectives of DRMS are to provide a set of interfaces for parallel environments that would facilitate the following:

• Allow application developers to construct parallel programs that are adaptable to run-time changes in available resources and are capable of spontaneous interactions with other applications, users, and the operating system.
• Allow users to cohesively manage one or more complex applications simultaneously and steer computations, using resources judiciously
• Provide system administrators with finer control over the resource allocation and consumption.

DRMS approach

• User specified points in a program where resource changes and interactions with external environment can take place.
• Provision of well-defined structure for expressing flexible data and control path.
• Parallel run-time system to interact with user applications and the operating system in a seamless manner.

Key features of DRMS

• Annotations to specify processor arrangements, data distributions, communication channels, and checkpoints.
• Run-time libraries provide application access to the information maintained by the run-time system.
• Interactive selection and processing of distributed data structures for visualization, communication, and post-processing.
• Inter-application communication for multidisciplinary problems
• Checkpointing, suspension, resumption, and migration of parallel processes.
• Application sensitive system administration facilities.
• Performance monitoring tools for application and system tuning.

Applicability

• DRMS is implemented on the IBM SP2.
• It is designed for parallel systems in general: scalable and distributed machines, symmetric multiprocessors, clusters of workstations.