Lionel Morel

Abstract - Monitoring Throughput Requirements of Streaming Applications Faced with Varying Execution Conditions

With the generalization of multicore and manycore chips, streaming languages are considered as a promising programming solution. They seem adequate for expressing many applications quite naturally and have been proven to be a good approach for taking advantage of the intrinsic parallelism of modern CPU architectures. Streaming programs are described as compositions of side-effect free actors that interact only through FIFO channels. Their expressivness ranges from static models which permit a high level of optimization at compile time to dynamic ones that allow to describe the most general types of programs. In this work we focus on static dataflow applications to be executed along-side legacy applications while satisfying quality-of-service requirements. We propose to monitor such programs at runtime both at the application and system level in order to precisely identify violations of quality-of-service requirements. Our monitoring relies on very little information from the programer and takes full benefit from the compilation of static dataflow. It allows to detect bottelnecks in the streaming graph and identify causes among cpu or memory overloading. The latter is of particular importance in the case of NUMA architectures. We also sketch how this monitoring can be used to dynamically adapt streaming programs faced with varying runtime conditions due to other applications running on the system.

Short bio

Lionel Morel received his MSc and PhD in Computer Science from Grenoble University, in France in 2001 and 2005 respectively. His initial research topics focused on programming models for critical embedded systems. After experiences at Abo Akademi in 2005-06 and INRIA-Rennes in 2006-07, he joined the Computer Science Departement and CITI lab of INSA de Lyon in 2007. There, he teaches processor architecture, operating systems concepts, concurrent programming and real-time operating systems. His research interests stem in Models of Computations for embedded systems, in particular dataflow programming, as well as programming models and operating systems support for energy-efficient embedded systems. Within the SOCRATE team he applies his research to softrware-defined radio and low-energy portable devices.