US 12,136,040 B2
Resource needs prediction in virtualized systems: generic proactive and self-adaptive solution
Souhila Benmakrelouf, Montreal (CA); and Nadjia Kara, Kirkland (CA)
Assigned to Telefonaktiebolaget LM Ericsson (Publ), Stockholm (SE)
Appl. No. 16/965,193
Filed by Telefonaktiebolaget LM Ericsson (publ), Stockholm (SE)
PCT Filed Feb. 5, 2019, PCT No. PCT/IB2019/050907
§ 371(c)(1), (2) Date Jul. 27, 2020,
PCT Pub. No. WO2019/150343, PCT Pub. Date Aug. 8, 2019.
Claims priority of provisional application 62/626,206, filed on Feb. 5, 2018.
Prior Publication US 2021/0064432 A1, Mar. 4, 2021
Int. Cl. G06F 9/50 (2006.01); G06N 3/086 (2023.01); G06N 3/126 (2023.01)
CPC G06N 3/086 (2013.01) [G06F 9/5077 (2013.01); G06N 3/126 (2013.01); G06F 2209/5019 (2013.01)] 14 Claims
OG exemplary drawing
 
1. A method for provisioning resources of an adaptive computing system in response to workload demands based on real-time prediction of future resource consumption by the adaptive computing system, the method comprising:
collecting resource consumption data (yi);
determining a real-time prediction of resource demand by the system comprising reading the collected resource consumption data (yi), initializing each of a size of a sliding window (ni) and a number of predicted data (mi) to a respective maximum such that (ni, mi):=(max ({ni}), max ({mi}));
setting an error-adjustment coefficient to minimize an estimated probability of a prediction error;
performing error adjustment on the predicted data based on the error—adjustment coefficient;
using a Genetic Algorithm (GA) to dynamically determine an optimal size of the sliding window and an optimal number of the predicted data within the real-time prediction of the resource demand to minimize the prediction errors;
adjusting the data within the real-time prediction of the resource demand based on an estimated probability of the prediction error and a variable padding, the variable padding based on a mean of at least one previous standard deviation of the predicted data within the real-time prediction of the resource demand calculated such that the variable padding=mean(σj(yi−ns,yi), where j∈{1, . . . , l}, l is the number of under-estimations greater than 10% and σj is the standard deviation of the jth under estimation;
after performing the error adjustment on the predicted data, based on the estimated probability of the prediction errors being underestimated, adding at least one padding value;
performing an initialization phase using the GA to dynamically determine the optimal size of the sliding window and the optimal number of the predicted data comprises determining an optimal pair (ns, ms) that comprises the optimal size of the sliding window and the optimal number of the predicted data;
using the optimal pair (ns, ms) to predict future resource consumption by the system based on the Kriging method and the adjustment of the prediction of the resource demand according to at least one error-adjustment value; and
based on the predicted future resource consumption, provisioning resources of the adaptive computing system in response to workload demands.