Supercomputing is a key enabling technology for most scientific disciplines: from physics to medicine and engineering. Thus, existing supercomputing infrastructures must attend a growing demand that can overpass the installed network capacity very easily. In this context, supercomputing centers usually must choose between two management policies: either assuming a loss in reliability and availability because of network congestion or implementing fair use policies to limit the free access to supercomputing resources. However, both have a negative impact on scientific production. Therefore, new innovative techniques to adapt the network resources in supercomputing centers to the demand at real time are required. In this paper we propose a solution to achieve a flexible network capacity in supercomputing centers, through time series prediction algorithms and ad-hoc 5G mobile communication interfaces. The individual traffic flow generated by every machine is predicted as a combination of three basic methods: polynomial fitting, time series decomposition, and Holt-Winters algorithm. This information is employed to detect future bottlenecks and congestion regions, using the teletraffic theory. When congestion is detected, an ad-hoc wireless 5G communication link is activated. A Particle Swarm optimization process is triggered to ensure the deployed 5G interfaces are technologically viable. An experimental validation based on simulation scenarios shows that supercomputing centers could support an increment in traffic up to 17%.
Supercomputing is a key enabling technology for most scientific disciplines: from physics to medicine and engineering. Thus, existing supercomputing infrastructures must attend a growing demand that can overpass the installed network capacity very easily. In this context, supercomputing centers usually must choose between two management policies: either assuming a loss in reliability and availability because of network congestion or implementing fair use policies to limit the free access to supercomputing resources. However, both have a negative impact on scientific production. Therefore, new innovative techniques to adapt the network resources in supercomputing centers to the demand at real time are required. In this paper we propose a solution to achieve a flexible network capacity in supercomputing centers, through time series prediction algorithms and ad-hoc 5G mobile communication interfaces. The individual traffic flow generated by every machine is predicted as a combination of three basic methods: polynomial fitting, time series decomposition, and Holt-Winters algorithm. This information is employed to detect future bottlenecks and congestion regions, using the teletraffic theory. When congestion is detected, an ad-hoc wireless 5G communication link is activated. A Particle Swarm optimization process is triggered to ensure the deployed 5G interfaces are technologically viable. An experimental validation based on simulation scenarios shows that supercomputing centers could support an increment in traffic up to 17%. Read More
Related posts:
Prediction of air traffic complexity through a dynamic complexity indicator and machine learning models
An inter-slice management solution for future virtualization-based 5G systems
A variation of the ATC work shift scheduling problem to deal with incidents at airport control centers
Initial Insights into Teleworking’s Effect on Air Quality in
Madrid City
An optimization algorithm for the efficient distribution of resources in 6G verticals
Determination of Air Traffic Complexity Most Influential Parameters Based on Machine Learning Models