Benchmarking of hard real-time distributed systems with Ada 95
Brian G. Ujvary, Nick I. Kamenoff, Jorge L. Díaz‐Herrera · ACM SIGAda Ada Letters · 1997
This paper presents an approach for the implementation of the Hartstone Distributed Benchmark (HDB) for Hard Real-Time Distributed Systems with Ada 95. From the series of experiments as defined in the original HDB paper, six experiments have been implemented for the sender-receiver (two node) version of experiments.The main idea behind the HDB is the definition of a number of (host processor) tasks on each node, where these tasks exchange messages with corresponding tasks on the other nodes. The stopping criteria for each experiment is the missing of a message before the deadline of a receiving task. The variables in the HDB experiments are: the number of tasks sending and/or receiving messages, the number of messages sent and/or received per task period, the length of messages sent and/or received per task period, and the number of nodes participating in the experiments.From the real-time scheduling point of view, the HDB experiments integrate the host processor scheduling domain (tasks' priorities), the communication scheduling domain (messages' priorities) both located on each node, and the inter-nodes communication domain (priority ordering among physical nodes). The host domain in the presented experiments is implemented by using the Ada 95 Tasking System. For the implementation of the communication exchange of messages, the Ada 95 Remote Procedure Call is used. The Network Time Protocol is used to provide the global time synchronization as required by HDB.The paper presents the design solutions, the actual implementation for the two nodes version, and the analysis of the results obtained with the current testbed. Design considerations for the implementation of the HDB experiments in the multi-node version, problems which can arise by using the Ada 95 Remote Procedure Call, as well as suggested solutions, are discussed.The results of the implemented experiments show a high level of repeatability and stability of the testbed setting used.