As service providers try to enable anytime, anywhere service models across any device, software-defined networking (SDN) and network functions virtualization (NFV) will be catalysts for service innovation including service mashups and service blending. But in order for operators to execute blended service models successfully, a policy-based and predictive analytics-driven approach to end-to-end service management will be essential. In the SDN and NFV world, balancing customer experience with resource optimization will be crucial because operators need to provide consistent, unified user experiences with QoS as the key differentiator for both virtual networks and physical services.
Predictive Analytics and Service Management
While most of the service management conversation focuses on service orchestration, the pivotal role that predictive analytics will play in service management is often overlooked. Next-generation service management platforms need to borrow principles typical to self-organizing networks (SON) and must feature self-healing capabilities. The purpose of this is to address device constraints or provisioning issues proactively and based on service quality before these issues impact the customer experience.
The dynamic nature of a next-generation, hybrid virtualized service environment demands self-monitoring for anomalous events in the network as well as the ability to diagnose and fix issues dynamically. This enables load balancing and optimizes network resources as well. Next-generation service management platforms must be policy-based and predictive-analytics driven to arm service providers with the ability to monitor and track real-time changes to the network at the per-application and per-subscriber levels while pre-empting service degradation by monitoring, measuring, and maintaining service metrics in real time.
An embedded predictive analytics layer should conduct “what if” analysis and forecast demand and workload allocation. It should also assign probability to future events and trigger pre-emptive actions. For example, if a Virtual Network Function (VNF) is exceeding design capacity, at what point should the VNF instance be scaled up? Or during periods of network congestion, when should the VNF move to a different data center and traffic be rerouted? Or when the cost of capacity delivery goes above a predefined threshold, at what point should VNFs be allocated to a lower cost solution?
Because the virtualized environment is inherently dynamic, operators must use big data and predictive analytics to collect and correlate the business metrics that drive decisions such as where to instantiate or move VNFs within the NFV infrastructure. Operators also want to understand where to instantiate VNFs based on business metrics, such as operating margin, server cost, or energy efficiency.
For enterprise customers, when there is a need for more bandwidth to maintain SLA levels or Quality of Experience (QoE) KPIs, predictive analytics will work closely with fulfillment and assurance systems to provide them with information pre-emptively. This way, additional virtual machines can be provisioned on the fly based on dynamic policies; the initial virtual machine can move to a higher capacity server or to one under less application load.
Making Service Delivery Smarter
With virtualization, service delivery requires service management based on policy management and predictive analytics, for a variety of reasons:
- When the service is first provisioned, the system must understand which resources to use and what the current loads are on impacted systems and networks
- Changes must be made live to provisioned services in order to maintain a positive customer experience
- Performance parameters must be measured in real time to help deliver customer-centric service experiences
- VNFs must be measured accurately to make sure that they can scale properly based on defined policies
- Provisioned policies must be monitored in real-time so that virtual machines (VMs) are allowed to move based on configuration policies
- Key parameters relevant in the virtualized world must be measured, including latency between network functions; performance of connecting links (VNF to VNF, VNF to legacy, etc.); bandwidth QoS; SLA measurement; intra-VNF component link measurement; SLA compliance management and verification; and measuring threshold alerts based on SLAs.
As NFV is adopted into mainstream production, VMs and VNF instances will start growing exponentially, making the management of such instances increasingly complex. The need to visualize and measure the performance of a complex and heterogeneous architecture will spread across multiple domains. As a result, the role of policy and predictive analytics will become mission critical in the context of holistic service management.