The configuration of a next generation 5G network is a very difficult task for the telecommunications operator or the Mobile Virtual Network Operators making use of the underlying telecom substrate. Many customer services require guarantees which should never be affected by the state of the substrate network and computing resources. Indeed, the commercial potential of many promising 5G applications will only come to life if the network itself is capable of automatically deciding and enforcing at runtime the most suitable resource allocations and configurations. Examples are many and range from 6LowPan sensor networks demanding new routing link estimators to improve performance and reduce delays to auto scaling of NFV resources implementing on demand IPTV services.
How the operator should allocate resources and configure them in an intelligent manner in order to cope with the enormous number of slices allocated to different services with different requirements?
Modio delivers an innovative solution for operators and IT software vendors for telecommunication networks. The heart of our innovation is the implementation of machine learning and classification capabilities and their integration with next generation management IT solutions for telco clouds. Our solution is customizable to be adopted to help solving real problems arising in a wide range of industrial use cases in wireless networking.
Core of our machine learning engine is the Qiqbus streaming analytics platform, an in house developed Spark based platform (see section Qiqbus) which Modio has been developing since 2012 and which has been installed in many commercial projects of Modio, for mobile advertising, Deep Packet Inspection and other commercial projects demanded by our industrial customers.
Architecture & Data Flow
A typical Qiqbus installation is depicted above. Monitoring data is pulled from input queues, grouped according to the source. The data is placed into the streaming computation engine for analysis. We employ a semi-supervised machine learning approach for the classification of computing and network metrics to specific performance categories. The computation engine (Spark) operates in micro-batching mode. It reads a group (batch) of events and forwards the group through a sequence of processing nodes. Events are aggregated according to configurable through the command line KPIs, and then clustered to distinct performance categories. A labelled set of existing metrics, collected through a bootstrap phase are often used as the source for training of the chosen performance classification model. The trained model is applied to the unlabelled classes and generates a weighed categorization of the contents of the unlabelled classes. The result is fed to a CEP engine, implemented in Spark streaming, which identifies whether an updated policy file should be sent to the management layer.
For more information on our 5G management technology contact us at firstname.lastname@example.org