Distributed Scheduling Aspects for Time-Critical Targeting
Project Award Date: 04-07-2003
KU researchers will collaborate with investigators from Washington University in St. Louis, the University of Rhode Island, and Ohio University in this DARPA project. The University of Kansas will serve as a subcontractor.
The project addresses methods for ensuring effective execution of computations whose components are distributed across a number of different physical components in a distributed real- time embedded (DRE) system.
An effective solution to this problem will require coordinated design and implementation at a number of different levels of the DRE systems, drawing on the expertise and previous work of all investigators. While each investigator has a primary area of expertise this project will also require a significant degree of effort from everyone to produce a fully integrated final product.
The University of Kansas will concentrate on two major research tasks, as well as participating in design, integration, and testing phases. The major research tasks addressed at KU will be the development of a System State Information Service (SSIS) and middleware supporting Group Scheduling (GS) of computation components.
The SSIS task will require researchers at KU to create methods for collecting, classifying, and disseminating information of interest to other system components as they perform their resource allocation and computation scheduling duties. The system state information will include the current use of system resources including CPU, memory, and network bandwidth. Classification activities are performed as close to the raw data as possible within the DRE architecture to minimize the communication resources required to support the SSIS. Dissemination of the state information will probably use a combination of push and pull semantics across a common application communication subsystem.
The GS task will address methods for scheduling computations implemented as groups of concurrent multi-threaded processes. This will require that we develop methods for the representation of the group of elements, processes or threads, implementing a computation, methods for controlling how resources are allocated within a group, and methods for controlling how resources are shared among many computations. The GS model will thus be hierarchic, permitting members of a group of elements implementing a computation to be a group as well. This task will also include significant experimentation with and evaluation of the range of methods by which the GS approach to the problem of scheduling computations can be supported by existing end-system operating system facilities.
The two major KU tasks will strongly interact with a major Washington University task of "addressing scenario-centric scheduling-related end-system mechanisms." The KU and Washington University teams will maintain an especially close collaboration to coordinate the design and implementation of these tasks.
The remaining work performed at KU as part of this project will consist of design and implementation coordination with other project tasks taking place at the other participating institutions, and tasks supporting the use of the Adaptable Performance Evaluation Testbed (APET) at KU.
All participants in the project will cooperate on the design of the system as a whole, but KU's effort in this will be particularly important because the information provided by the SSIS will be used by components developed by all other participants. The group scheduling methods developed by KU will provide basic descriptive and control capabilities required by other project components to impose the desired control over distributed computations. Part of the work at KU will thus be maintaining close coordination of design and implementation with the different parts of the project.
The APET will be used for large-scale tests of the methods and implementations developed by this project. The KU APET task will include configuration of the APET for use by the project. This will include installation and integration of the project test environment components on the APET and ensuring that the APET facilities allowing remote access are adequate to support the project testing activities by project participants.
Faculty Investigator(s): Douglas Niehaus (PI)
Student Investigator(s): Tejasvi Aswathanarayana, Karthikeyan Varadarajan, Deepti Mokkapati, Hariharan Subramanian, Sriram Chadalavada, Michael Frisbie, Noah Watkins, Andrew Boie, Ashwin Kumar Chimata
Primary Sponsor(s): DARPA