At TM Forum Live! (May 15-18, Nice, France), Appledore Research’s Patrick Kelly will give a presentation on ‘Automated assurance for cloud services’. He provides an introduction here.
The telecommunications industry is amid a major transformation both in terms of business models and technology. Virtualization, cloud computing, and 5G will be significant technology transformation drivers creating entirely new markets. Virtualization, and the automation it enables, is rapidly changing how we assure communication services, provide reliability and availability, optimize capacity and utilization, deliver a superior customer experience, and generate new sources of revenue and customer value. Programmable networks abstract the network intelligence into the software layer and therefore require a new management and operational paradigm to supporting advanced services.
RASA: An innovative approach to assuring services
Appledore Research Group introduced a functional architecture that is designed to improve the operating efficiencies and service lifecycle management of cloud-enabled services. Furthermore, we felt it was critical to re-imagine service assurance such that it powered efficient control loops, using existing best practices. We call this architecture Rapid Automated Service Assurance (RASA). It addresses the need to manage both the existing physical infrastructure and the virtualized network functions that will serve as the underlying resources to provide dynamic services in the digital economy.
It is our view that existing OSS systems — which are structured as dual software stacks implementing fulfillment and assurance — must merge to facilitate automated workflow processes. The cloud is on-demand, elastic, ubiquitous, and resilient. Figure 1 is a simplified architecture view that shows the interaction between both management operational functions and the ability to automate and close the loop.
To construct a model of the underlying network and resources, data must be acquired by a discovery that is in a perpetual state of change. Dynamic changes to the resources and topology are maintained by inventory, which essentially supports other software functions in the assurance and orchestration domain.
We like to think of the dynamic inventory and the live topology as a shared resource. Subsets of the dynamic inventory may exist in the RASA or orchestrators and this data must be federated to avoid unsynchronized data sets.
The role of RASA is to process a “live” stream of user plane and control plane performance data, and compare active streams to a signature pattern for the service.
After applying advanced analytics to the measured data, the system will isolate the service-impacting event and generate a notification, which can be automated via a policy-driven logic. At this point the RASA system is providing the understanding which will drive corrective action (scaling, healing, moving) by the appropriate orchestration system/layer. Any reconfiguration, instantiation of VNFs (virtualized network functions) or re-routing of network paths will be initiated at this point. The process will then be updated in the discovery to inventory to change in topology state. This activity occurs in very short cycle times to enable RASA to analyze and then trigger any future state changes completing the control loop.
As networks become more complex and multi-domain, we believe that it will be increasingly difficult to anticipate all possible sources of impairment: consequently, we will need systems capable of learning them. The subject of machine learning has been applied successfully to other industries but has yet to receive the wide appeal in predicting future service impact in the telecommunication domain.
Google has used it successfully in their mapping technology to offer suggestions of re-routing a driver as traffic patterns change based on accidents and weather. It is being applied to self-driving automobiles and in the health science field to provide better healthcare.
We think machine learning is beneficial in understanding the dynamic nature of virtual networks. Signatures have been used in the past to accelerate problem resolution. It assumes a base of data to begin the analysis. Currently we are looking at early attempts to use machine learning in a supervised learning mode to accelerate the understanding process in a dynamic virtualized infrastructure and then utilize this method to drive automation in the orchestration and policy domain layers.
The market for commercial service assurance solutions is maturing from the “customer experience management” era towards the era of “closed loop automation” (figure 2)
The closed loop automation era will be driven by the needs of truly dynamic cloud architectures, virtualization and the business requirements necessary to deliver on-demand services, proactive healing and dynamic sharing of costly infrastructure. We advocate that increased automation is the only way forward to support the demands of cloud services. And service assurance must be proactive if CSPs want to improve customer experience. Without automation, many of the economic benefits of virtualization will be foregone, and at the same time, the added complexity may well result in worse customer satisfaction, and a greater operational burden – meaning rising, not falling costs. It is critical then that CSPs design for automation and for scale, and reap both lower costs and greater flexibility.
TM Forum Live! takes place May 15-18 in Nice, France. Find out more at www.tmforumlive.org