Resource optimization for CSDF-modeled streaming applications with latency constraints

Abstract

Abstract-In this paper, we study the problem of minimizing the number of processors required for scheduling latencyconstrained streaming applications modeled as CSDF graphs, where the actors of a CSDF are executed as strictly periodic tasks. We formalize the problem and prove that due to the strict periodicity of actors the problem is an integer convex programming problem, that can be solved efficiently by using an existing convex programming solver. We evaluate our solution approach on a set of 13 real-life streaming applications modeled as CSDF graphs and demonstrate that it can reduce the number of processors in more than 52% of the conducted experiments in comparison to an existing approach. I. IntroductionStreaming applications, such as video/audio processing and digital signal processing, have become prevalent in embedded systems. These applications contain ample amount of parallelism which perfectly matches the processing power of MultiProcessor System-on-Chip (MPSoC) platforms. To efficiently program MPSoC platforms, Models-of-Computation (MoCs) are used to specify streaming applications. Prominent examples of MoCs include Synchronous Data Flow (SDF) [1] and its generalization Cyclo-Static Dataflow (CSDF) [2], in which actors representing computation are executed concurrently, thereby naturally exposing parallelism. Furthermore, the strong design-time analyzability of these MoCs makes them suitable for designing performance-constrained embedded systems.Performance constraints of a streaming application are usually imposed on two principle metrics, throughput and latency. For many streaming applications, the latency is the main concern, where latency is the elapsed time between the arrival of a sample to an application and the output of the processed sample by the application. For instance, in video conferencing and automatic pattern recognition applications, a latency that exceeds a certain limit cannot be tolerated. At the same time, the required number of processors to execute the application should be minimized for better resource usage and energy efficiency.A few efforts have been made to deal with latency of streaming applications specified as SDF/CSDF graphs. The authors in [3] studied minimizing latency for SDF graphs, where latency is computed by a state-space traversal which has exponential complexity. Moreover, they assumed that there is no constraint on the number of processors required to schedule an application. However, the number of processors is an important design concern for embedded systems with respect to power consumption and area. In another effort, the authors in [4] and [5] proposed a scheduling framework that schedules acyclic CSDF graphs in a strictly periodic fashion. In this scheduling framework, each CSDF actor executes strictly periodically and meets a given deadline. The periodic execution of actors guarantees a certain