An optimization approach for dynamic allocation on reconfigurable resources at runtime

摘要

Reconfigurable computing can be divided into two types including static reconfiguration and dynamic reconfiguration according to the reconfiguration timing. Static reconfiguration can provide low runtime overhead for the precomputing configuration. Due to the failure to know the run-time information of the system, once the static prediction is different to the real profiling, static scheduling cannot obtain high performance without flexible. While dynamic scheduling can be used to schedule decisions according to the runtime information of the system. It will dynamically allocate the reconfigurable resources to the tasks at runtime, which can provide high performance and considerable flexibility. However, dynamic allocation at runtime will also have some overhead for the instant resource management. This paper presents an optimization approach for dynamic allocation on reconfigurable resources at runtime. Hardware tasks can obtain available reconfigurable resources according to its resource request and the system profiling at runtime. The experimental results show that this approach can provide better performance than traditional ones.