Intelligent resource allocation for 5G cloud environments by Modio Resource allocation within NFV cloud substrates requires high dynamism; static mechanisms cannot cope with large variations of resource demand without avoiding either over allocation of resources that leads to excessive operational costs or under allocation that leads to SLA violations of the offered services. To solve this problem, Modio’s experiment, Intelligent resource allocation for 5G cloud environments, implemented a novel computational intelligent-based approach to dynamically assist the dynamic resource allocation mechanism, implemented within OpenBaton’s autoscaler. Specifically, Modio implemented a predictive agent incorporating state of the art forecasting algorithms (ARIMA, Holt Winters and Recurrent Neural Networks) in order to predict the upcoming system load and dynamically generate the autoscaling policy content. Overall, Modio experiment a) implemented an autoscaling policy agent which uses thee different forecasting methods; b) integrated the policy agent with OpenBaton’s autoscaling engine; and, c) performed series of experiments to compare the performance between: A) pre-configured policies in OpenBaton where action parameters are manually configured; and, B) the machine learning approach for dynamically generating the parameters of scale-out and scale-in actions. This experiment employed SoftFIRE’s testbed resources and Modio conducted series of experiments in order to demonstrate the performance of the approach for a WebRTC service and measured it against the KPIs that were defined in collaboration with the SoftFIRE consortium. Experimental results demonstrated that the forecasting models implemented in Modio predictive agent are effective in deriving appropriate scale-out and scale-in actions’ parameters which outperform OpenBaton’s autoscaling policy with fixed step. Especially for the third KPI defined in the experiment -which tackles the problem of over provisioning of resources – the experiment has shown how the forecasting models can help reducing the number of active VNFs when the input load is decreasing, while also saving energy for the 5G cloud provider by using less CPU cycles and, at the same time, increase the number of SLA conformant clients. From Modio business point of view, this project has provided important technical knowledge in experimenting with the OpenBaton NFVO and the SoftFIRE testbed. The results obtained have provided with confidence that Modio concept has a promising potential for exploitation. However, there is room for improvement of the work, before it can reach the stage of a mature Minimum Valuable Product (MVP) which Modio can showcase to potential VCs and/or vendors who may integrate the system to enhance their portfolio. On the other hand, our implementation has the potential to result to an extension of OpenBaton’s autoscaler, enhancing it with predictive functionality. Concluding, this experiment allowed Modio to validate a Proof of Concept implementation. It also encouraged Modio to actively pursue collaborations with relevant stakeholders and invest resources and/or actively purse VC backing for continuing the work with the goal to market the technology. Clearly, the results obtained by the experiment validate that Modio concept has chances for business exploitation. The challenge is to reach the right collaborators and target the relevant potential adopters of the technology. To that end, Modio would like to further collaborate with the SoftFIRE consortium in order to gain more knowledge on business plans related to NFVO’s and their individual features. In parallel, Modio will disseminate its work through inbound and outbound marketing channels, Modio will participate in relevant industrial fora and will publish the results obtained in this experiment, aiming to attract the attention of potential adopters of the technology. Listen to the Modio interview Watch the Demo of Intelligent resource allocation for 5G cloud environments Experiment
