Conflict-free networks on chip for real time systems
- Autores: Tomás Picornell Sanjuan
- Directores de la Tesis: Carles Hernández Luz (dir. tes.), Jose Flich Cardo (dir. tes.)
- Lectura: En la Universitat Politècnica de València ( España ) en 2021
- Idioma: español
- Tribunal Calificador de la Tesis: Pedro Juan López Rodríguez (presid.), José Cano Reyes (secret.), Giorgos Dimitrakopoulos (voc.)
- Programa de doctorado: Programa de Doctorado en Informática por la Universitat Politècnica de València
- Materias:
- Enlaces
- Tesis en acceso abierto en: RiuNet
- Resumen
The ever need for higher performance to cope with the high computational power demands of new applications (e.g autonomous driving systems), forces industry to support technology based on multi-processors system on chip (MPSoCs) in their safety-critical embedded systems. MPSoCs usually include a network-on-chip (NoC) to interconnect the cores between them and, with memory and the rest of shared resources. Unfortunately, the inclusion of NoCs difficults achieving time predictability as network-level conflicts may occur in many points in a distributed manner.
To overcome this problem, this thesis proposes a new time-predictable NoC design paradigm where conflicts within the network are eliminated by design. This new paradigm builds on top of the Channel Dependency Graph (CDG) in order to deterministically avoid network conflicts. Our solution is able to naturally inject messages using a TDM period equal to the optimal theoretical bound without the need of using a computationally demanding offline process. The network is integrated in a tile-based manycore system and adapted to its memory hierarchy.
As a second main contribution, we propose a novel distributed dynamic scheduler that is able to achieve peak performance close to a wormhole-based NoC design without compromising its real-time guarantees. The scheduler builds on top of our NoC design to exploit its key properties.
The results of our NoC show that our design guarantees time predictability avoiding network interference among multiple running applications. The network always guarantees performance and also improves wormhole performance in a 4 x 4 setting by a factor of 3.7x when interference traffic is injected. For a 8 x 8 network differences are even larger. In addition, the network obtains a total area saving of 10.79% over a standard wormhole implementation.
The proposed scheduler achieves an overall throughput improvement of 6.9x and 14.4x over a baseline conflict-free NoC for 16 and 64-node meshes, respectively. When compared against a standard wormhole router 95% of its network throughput is preserved while strict timing predictability is kept. This achievement opens the door to new high performance time predictable NoC designs.
As a final contribution, we build a taxonomy of TDM-based NoCs with real-time properties. With this taxonomy we perform a comprehensive analysis to study and compare from response time specific, to low resource implementation cost, through trade-off solutions for real-time NoCs designs. As a result, we derive new TDM-based NoC designs.
