Angol nyelvű Konferenciaközlemény (Könyvrészlet) Tudományos

Megjelent: IEEE. NOMS 2020 - 2020 IEEE/IFIP Network Operations and Management Symposium: Management
in the Age of Softwarization and Artificial Intelligence. (2020) ISBN:9781728149738 Paper: 9110387

Azonosítók

- MTMT: 32698410
- DOI: 10.1109/NOMS47738.2020.9110387
- WoS: 000716920500112
- Scopus: 85086767719

Decoupling congestion control plane from datapath can expedite the development of
new congestion control solutions. It also creates opportunities for explicit rate
allocation schemes. Dealing with large numbers of flows remains a major challenge.
Max-min fairness - the gold standard for flow rate allocation has a running complexity
proportional to the number of flows, which might be prohibitive in large-scale networks.To
accelerate explicit rate allocation, we present solutions using rate quantization,
i.e. mapping the continuous range of flow rates to a small number of bins. We use
Lloyd-Max, a quantization method that generates bins according to the distribution
of flow rates, to dynamically adjust the quantization bins over time. Our experimental
evaluation shows that the distortion caused by this quantization scheme is small,
and can be negligible compared to intrinsic errors in measuring and enforcing rates
in current solutions.We also show that rate quantization can significantly speed up
max-min fair rate allocation, reducing the run-time by 70 95%. Besides, Lloyd-Max
quantization using recent history of flow rates performs close to the case when we
have access to the exact current (or future) rates. This is an interesting observation
as it obviates the need for complex techniques that try to predict future rates.

2024-11-03 10:18