QoS based Resource Allocation Model

in Dynamic Real-Time Systems

(DQRAM)

Abstract:

Dynamic real-time systems require adaptive resource management to accommodate varying processing needs. We address the problem of resource management with a single resources for soft real-time systems (no hard deadline requirements) consisting of tasks that have discrete QoS (Quality of Service) settings that correspond to varying resource usage and varying utility (benefit to the end user). Given an amount of available resource, the problem is to provide on-line control of the tasks’ QoS settings so as to optimize the overall utility. Since the complexity of the problem precludes optimal solutions, we present a discrete control theory that can be implemented with a heuristic control algorithm with the following properties.

(1) It has low runtime complexity, making it suitable for an on-line system.

(2)It makes incremental adjustments to QoS settings as available resources change, avoiding the calculation time and instability that would be incurred by recalculating all the QoS settings.

(3) Differences between actual utility and optimal do not accumulate over time, so there is no long-term degradation in performance.

(4) The lower bound on actual utility can be calculated dynamically based on current system conditions, and an absolute lower bound can be calculated statically in advance.

(5) It uses feedback mechanisms to respond to actual resource availability, allowing all resources to be used and tolerating misspecification of task resource requirements.

complete paper submitted to ACC available here....

About ' drt-qos-man' Project:

The ' drt-qos-man' aims at implementing, testing and verifying  algorithms A2 and A3 discussed in the ACC Paper. This research project represents the QoS Manager from the proposed architecture. The project will be further expanded to include the  Allocation Manager (which may be envisioned as algorithms A4 and A5 ).

' drt-qos-man' project currently provides a 'C' library, libQoSMan, which may be considered as a reference implementation of QoS MAnager. A simulation is provided which demonstrates how the algorithms behave as compared to the exhaustive search. This simulation uses the libQoSMan.

libQoSMan is the first step towards development of a kernel module for integration of  QoS Manager into the Linux kernel.

A discrete Event Simulation for QoS Manager is currently being written using OMNeT++ tool and libQoSMan. Bleeding edge source for this simulation can be obtained from the cvs repository from here or .

For More info about this project, Please contact : Jayram A. Deshpande
Copyright ©2004 CIDDS@ Ohio University