Energy-, Latency and Resilience-aware Networking
Final Report Abstract
e.LARN has continued and completed the work done in LARN. We added energy-constraints to the design of our Reliable Networking Atom (RNA) and achieved as well as published several new and partially surprising findings: 1. We introduced and leveraged X-Leep, a platform to measure and document as well timing- as energy-related statistics. 2. We derived and implemented a near-optimum DVFS (dynamic voltage and frequency scaling) policy and documented that we could significantly reduce the energy consumption of our RNA without sacrificing the latency and resilience predictability. 3. We could show that the design criteria for error coding on the transport layer are completely different from those on the physical layer: The complexity of decoding is not decisive, since the required matrix inversion only needs to be done once but is applied many hundred times per packet. Hence the encoding and decoding complexity grows linearly with the block length. 4. Despite the only linear growth we could show that binary codes due to their very simple arithmetic operations (XOR and AND) can be advantageous if block lengths are sufficiently large (≥≈ 25). 5. We investigated on the multicast-capability of binary codes and could leverage the persymmetric structure of Polar-codes (as e. g. used in 5G) to create Polar-codes with excellent multicast capabilities. We also derived bounds for binary codes optimized for multicast (MC-FEC ). 6. We leveraged neural networks / machine learning for two important ingredients of cyberphysical networks: a) the parameterization of the HARQ (Hybrid Automatic Repeat reQuest), where we could show that the inference time is very well predictable, i. e. it has a much smaller variance than our greedy search and b) light weight online machine-learning that significantly improves the system-level timing. The results sketched above led to 15 publications in organs with scientific quality assurance.
Publications
-
Precisely Timed Task Execution. 2020 IEEE 23rd International Symposium on Real-Time Distributed Computing (ISORC) (c(2020, 5)), 10-19. IEEE.
Reif, Stefan & Schroder-Preikschat, Wolfgang
-
X-Leep. Proceedings of the Eleventh ACM International Conference on Future Energy Systems (c(2020, 6, 12)), 548-553. ACM.
Reif, Stefan; Herzog, Benedict; Pereira, Pablo Gil; Schmidt, Andreas; Büttner, Tobias; Hönig, Timo; Schröder-Preikschat, Wolfgang & Herfet, Thorsten
-
Reducing FEC-Complexity in Cross-Layer Predictable Data Communication. 2021 IEEE 18th Annual Consumer Communications & Networking Conference (CCNC) (c(2021, 1, 9)), 1-2. IEEE.
Pereira, Pablo Gil & Herfet, Thorsten
-
Polar Coding for Efficient Transport Layer Multicast. 2022 IEEE 19th Annual Consumer Communications & Networking Conference (CCNC) (c(2022, 1, 8)), 313-318. IEEE.
Pereira, Pablo Gil & Herfet, Thorsten
-
Resource-demand Estimation for Edge Tensor Processing Units. ACM Transactions on Embedded Computing Systems, 21(5), 1-24.
Herzog, Benedict; Reif, Stefan; Hemp, Judith; Hönig, Timo & Schröder-Preikschat, Wolfgang
