A new scalable service discipline for real-time traffic: The framed-deadline scheduler
Computer Communications
| Hi-index | 0.00 |
Clusters of workstations are being extensively used forsolving computationally intensive scientific problems.However, there is limited support for Quality of Service(QoS) based distributed computing on commercial off-the-shelf (COTS) clusters. This limitation has restricted successful deployment of distributed real-time high-performancecomputing applications to customized anddedicated embedded multi-processor systems. This paperdescribes research work that attempts to provide a clusterplatform that can guarantee access to computational andcommunication resources to distributed applications. Theauthors have developed PromisQoS, an architecture thatsupports execution of hard real-time distributedapplications on a Linux cluster while providing high-throughputand low-latency communication usingMyrinet. PromisQoS consists of the following majorcomponents -- Hare, BDM-RT and Turtle. Hare is aprototype implementation of time-based QoS channelsspecified by the Real-Time Message Passing Interface(MPI/RT 1.1) standard. BDM-RT is a low-level messaginglibrary on Myrinet that provides deterministiccommunication latency and bandwidth on Myrinet. Turtle,a variant of RT-Linux, is the real-time operating systemthat provides guaranteed computation time. This workdemonstrates that it is possible to deploy hard real-timedistributed applications on COTS clusters and underlinesthe significance of the MPI/RT API in the realm ofdistributed high-performance computing applications thatrequire QoS.