The need to increase the performance of real time systems is growing along with system complexity. High performance computers (HPCs) with real-time scheduling support can be used to control and improve the performance of real time engineering applications. The latency that develops when parallel programs finish at dissimilar times is referred to as jitter. Jitter and latency can develop due to interference by other processes, interrupt handlers, or the Linux operating system. Experiments that used the Real Time Application Interface (RTAI) in conjunction with the Message Passing Interface (MPI) to implement parallel applications, reduced or eliminated jitter for experimental codes that have characteristics typical of engineering applications. The experimental HPC test bed is a Linux cluster with nine Intel Pentium IV, 3.4 GHz computers, connected by 100 Mb Ethernet using a switch. Each Linux system has 1 GB main memory, and is running Linux release 2.6.23 patched with RTAI 3.6.
A multi-DAG model for real-time parallel applications with conditional execution