A component-based programming model and support middleware for hierarchical real-time fault-tolerant computing in wide-area network environments.

摘要

Distributed computing in wide area networks (WANs) has become an active research field in recent years, especially under the label of grid computing. However, the field of real-time distributed computing (RTDC) in WAN environments has remained in an embryonic stage. The need for significant advances in this area is becoming more important as demands for new-generation RTDC applications increase. Examples of new-generation RTDC applications are wide-area multi-party videoconferencing, collaborative remote surgery, collaborating robots, and freeway auto-navigation systems.; The major research challenge in WAN-based RTDC involves two aspects: (1) reliable message communication among widely distributed nodes with easily determinable tight latency bounds, and (2) allocation of computing and communication resources toward the effect of application services occurring without violating timing requirements. Moreover, an important obstacle in developing easily analyzable and highly reliable RTDC applications in WAN environments is the lack of an adequate programming model.; This dissertation work presents a component-based programming model and support middleware architecture for hierarchical RTDC in WAN environments. The proposed programming model and the support middleware are extensions of the Time-triggered Message-triggered Object (TMO) programming scheme and the TMO Support Middleware (TMOSM), originally established for safety-critical RTDC applications in local area networks (LANs).; One of the major techniques introduced in this work is a hierarchical TMO-region-based wide-area RTDC programming model that relieves WAN-based RTDC application designers and programmers from dealing with complicated resource management and varying delays among computing nodes.; This work also presents the design and implementation of a real-time resource manager responsible for establishing a TMO-region network and allocating resources based on application requirements. The core element in resource management is the region-based distance-aware resource allocation algorithm, which achieves effective resource allocation according to the application requirements. The algorithm minimizes inter-region communication traffic to improve performance efficiencies in wide-area network environments.; Furthermore, new features for supporting WAN-based RTDC applications are incorporated into the existing TMO Network Configuration Manager (TNCM) of TMOSM. The augmented TNCM is called TMO Application Manager (TMOAM) and includes: (1) a mechanism for dynamic registration and deployment of real-time computing components (TMOs) with determinable latency, (2) A run-time TMO discovery mechanism with determinable latency, and (3) a zero-latency primary-shadow-mode-based TMO migration mechanism. The mechanism guarantees that at least one instance of the migrating TMO is active and generates correct outputs for other TMOs. The migration procedure does not affect the execution of other TMOs. Hence, from the other TMOs' viewpoint, there is no latency in the migration procedure.; In addition, a new scheduling scheme has been designed and implemented to achieve low scheduling latency and high CPU utilization rate in dual-core and multi-core computing platforms. Experimental results show significant improvements on dispatching latencies for application threads using the new scheduling policy.; Finally, the last part of this dissertation work describes a hierarchical wide-area fault-tolerant support framework, an extension of the LAN-based supervisor-based network surveillance scheme of the TMOSM. The fault-tolerant framework provides an enhancement to WAN-based RTDC middleware with fault-detection mechanisms in wide-area network environments.