| The Role of Network Tra c Statistics in Devising Object Migration Policies (2008) | |||||||||||||||||
Abstract | |||||||||||||||||
| To achieve the goal of improving performance, reliability, and concurrency control in our real{ time groupware system DISCIPLE, we are designing a knowledge{based system for resource control and management. An important part of our strategy is ne{grained resource control in terms of managing object location. Objects migrate to di erent hosts according to migration policies to accomplish their tasks. Our goal is to develop a mechanism for run{time learning of migration policies which is transparent to the user as well as to the application programmer. This paper addresses the role of network tra c statistics in learning object migration policies for real{time groupware applications. 1 The DISCIPLE Real{Time Groupware System As a part of the real{time collaborative groupware project DISCIPLE,we are investigating the role of network tra c statistics in resource management. The DISCIPLE system at CAIP is an advanced groupware design that enables multiple participants, using networked computers at di erent locations, to collaboratively access, manipulate, analyze, and evaluate multimedia data. The system uses knowledge{based planning and learning strategies for discerning the communication needs of participants and computational task demands. The goal of DISCIPLE is to provide software reliability, quality of service, and concurrency control in a distributed environment. To achieve this goal, we are developing knowledge{based systems for management and control of various system resources. One such knowledge{based system manages and controls object location in a distributed environment. This system moderates the work of an object communication infrastructure (CORBA{compliant Object Request Broker [5]) and its functioning would ideally be transparent to the user and to the application programmer (Figure 1). | |||||||||||||||||
Publication details | |||||||||||||||||
| |||||||||||||||||