General Purpose Process Migration

Arizona State University, Scottsdale AZ

Investigators

Abstract

PI: Partha Dasgupta The process migration problem has received a lot of attention but has produced only mixed results. Traditionally, migration mechanisms have been used to balance the load among processors in a distributed system. However, most mechanisms proposed to date for general-purpose applications have required extensive operating system kernel modifications. Hence, conventional wisdom has held that the costs of process migration are not justified by resulting benefits. This research takes a new look at process migration, recognizing both its utility for an increasing number of emerging applications that can tolerate higher overheads, as well as the possibility of implementing necessary mechanisms using novel user-level techniques. Process migration enables a host of capabilities such as mobility, collaborative work, distributed systems management, automatic reconfigurability, and fault-tolerance. Moreover, API interception technologies, which permit insertion of arbitrary functionality in the execution path of an application's interactions with the underlying operating system, enable process migration to be implemented without kernel modification. This research investigates a virtualization scheme that decouples an application from dependencies in the operating system and the physical environment. Virtualization also monitors and controls the interactions and the resulting side effects between processes and operating systems, making process migration possible.