Pranav Tendulkar, Peter Poplavko, Jules Maselbas,
Ioannis Galanommatis, and Oded Maler
A Runtime Environment for Real-time Streaming Applications on Clustered Multi-cores (2015)
A Runtime Environment for Real-time Streaming Applications on Clustered Multi-cores (2015)
TR-2015-6.pdf
Keywords: multi-rate dataflow graph, middleware, DMA, scheduling, schedulability, real-time systems, data parallelism, multi-core compiler, multi-core programming model, multiprocessor network-on-chip, many-core platforms, multiprocessor application benchmarks, multimedia, video and image processing
Abstract: Dataflow graphs with static rates, such as synchronous and cyclo-static dataflow models, can be used for multi-core programming of so-called streaming applications, which include a wide range of signal processing and coding. Given soft and firm real-time constraints on throughput, buffer size, and latency, many scheduling optimizers have been proposed for mapping dataflow models on multi-cores. However, not many of them are equipped with runtime environments that can deploy and execute the optimized schedule on the target platform. Among the publicly available runtimes, one can hardly find any that inherently support clustered multi-cores. Those are on-chip multiprocessors that consist of clusters, i.e. groups of multiple processors that share a local memory with each other and communicate via a global network-on-chip. This is an important possibility to ensure scalability in the number of cores, making it possible to provide a large (more than 100) number of processors on a single chip. Along with our StreamExplorer open-source scheduling optimizer for static dataflow models we propose a runtime environment that deploys optimized scheduling solutions on a clustered multi-core platform.
