Performance analysis of the Disrupted Static Priority scheduling for AFDX
Résumé
The AFDX technology is used as backbone in sev-eral aircraft. It offers a high bandwidth (commonly 100Mb/s), and guaranteed per data flow a bound on the network traversal delay, while being a shared resource. To do so, it uses a segregation mechanism, the Virtual Link (VL), designed as a bus abstraction. Nevertheless, there may exist applications with such stringent latency requirements that a given AFDX topology can not ensure it. To provide low latency, the AFDX technology provides priority levels: but even in the highest priority level, the latency bound can still be too high with regard to applicative requirements. One cause of latency is the non-preemptive aspect of priority policy: when a high priority frame reaches an output port, it may have to wait the emission of a low priority frame. Then, a high priority frame might have to face a delay equal to the maximum packet emission time of the low priority flows per crossed switch. The scheduling policy called Disrupted Static Priority (D-SP) aims to improve the latency guarantees, by interrupting low priority frames, and re-emitting them from the beginning. This policy is a trade-off between simplicity and efficiency. A quantitative comparison between the NP-SP and the D-SP scheduling systems' performance on a realistic case-study is provided in this article.
Origine : Fichiers produits par l'(les) auteur(s)