As CSPs begin the journey of becoming data-centric, they may find that building a full analytics stack that is carrier-grade, manageable, and performs at the scale required to analyze the volume of data necessary is very challenging. Each step requires significant investment. From hiring big data engineers and data scientists with the expertise needed, to developing the appropriate algorithms and data models, all are monumental tasks.
Airlines don’t make their own planes
I once heard someone say, “There is a reason airlines don’t make their own airplanes”. If every airline did this, they would have to develop advanced metallurgy and manufacturing processes while simultaneously optimizing the operational aircraft, including things such as engine management, environmental management, flight control systems and much more. They would also have to accept the full cost of repairs, failures, improvements, and optimization. This is simply too expensive and is not the core of their business.
The airline’s business is transporting people and goods. They may differentiate themselves with custom interiors, amenities, price, and scheduling, but not with the construction of the plane itself. A plane type and model can be used by any airline. This allows the cost of research, development, warranties, and safety improvements to be prorated across all airlines who use the same type of plane. The airline benefits from having the best technology available and differentiates itself with the services it offers inside the plane itself.
CSPs do the same by purchasing network equipment from network equipment providers and differentiate themselves through the design of the network and its operation. Similarly, pairing up with a full-stack analytics partner can accelerate the analytics journey and let CSPs differentiate themselves through the optimization of their operations and customer experience.
Telecom analytics framework
The telecom analytics framework is very different from an enterprise analytics solution. For one, collecting trillions of records per day from your Radio Access Network and fusing them with electronic data records from gateways to generate a policy that maximizes subscriber QoE is a pretty specialized task. Now factor in that a cell site may be predicted to congest within 15 minutes and you have a multi-faceted analytics problem. The sheer volume of data is mind boggling and yet the time-value of much of this data is very short. This implies that writing the data to disk in your data warehouse may not be economical.
As an example, compute-first architectures may be employed to reduce the cost of analysis. Application aware compute-first architectures utilize intelligent collectors that perform a data triage process, so the high-value portions of the data stream are collected and analyzed while the remaining data is discarded. In distributed networks this is often performed at the point of collection. This reduces the cost of transport and storage while speeding the analytics process, providing insights at the right time for the right cost.
In traditional streaming architectures, collecting data from these distributed portions of the network often requires a large amount of transport capacity to move the data from the source to the analytics platform. Transport and storage has often been cost prohibitive and a key limiting factor in deploying analytics on a broad scale across CSPs. The ability to collect data at the network edge while simultaneously reducing the data in a lossless and real-time way, is critical in making use of data that was too costly to collect and transport in the past.
The industry is moving faster than one can even imagine with the number of open source and commercial big data technologies available growing by the day. Within this maze of technologies, how do you select the right ones for your business? The complexity of selecting the right technologies and integrating them all while ensuring the best compute efficiency, scaling, and feature richness can be overwhelming. Add to this the need to maintain a telecom-grade management system with full fault-tolerance, capacity, alarm, and performance management capabilities and you have a significant challenge. Is it impossible? Far from it. But is it the best use of your time?
Choose a partner that brings years of know-how and expertise in combining big data environments with analytics to the table. Understanding how to optimally blend open-source and industry-specific algorithms and proprietary machine learning algorithms in a production environment is no trivial task and cannot be accomplished overnight. You need a partner who has experience deploying CSP applications in production environments and has the domain expertise to co-innovate with you, allowing your developers to focus on creating applications that are tailored to your business needs. This enables your developers and big data teams to drive the most value from your core business in the least amount of time, whether that be flying an airplane or providing service to millions of subscribers.
Image attributions: KaterynaDi & MarchCattle/Bigstockphoto.com