Introduction

Next-gen mobile networks have a unique opportunity to deliver powerful services that take advantage of low-latency to devices, high bandwidth networks, and edge computing to deliver services that are immersively integrated into the edge environment. This piece describes how Swim provides everything providers need to deliver new services using Multi-access Edge Computing (MEC).

• MEC can allow providers to offer new services and host customer applications that depend on real-time insights into the states of, and relationships between, devices in the edge context.
• It can also give providers real-time insights that enable them to optimize service delivery and dynamically control their core networks to maximize customer quality, enable slicing and QoS, and quickly find faults.

MEC offers low latency computing co-located with wireless base-stations and close to fixed networks. It interfaces directly to the Radio Access Network (RAN), which manages connectivity to wireless devices. MEC environments are richly connected - to the provider core network, the Internet, and public and private clouds.  As a result, MEC offers a unique opportunity to providers to deliver valuable services that use continuous intelligence to analyze, learn & predict from streaming data, on-the-fly.

Continuous intelligence applications can integrate with public and private cloud services. Providers can avoid data custody challenges by taking advantage of an “analyze, react (and then store)” architecture in which valuable, low latency responses are computed on-the-fly in the MEC context, but data storage occurs subsequently, typically in a public cloud or on-prem. This simplifies application architecture since data is not stored in the MEC environment.

Mobile + Edge Computing = $

What do providers need to turn MEC into a powerful revenue generation platform for services and applications that can deliver insights a million times faster, and at a tenth of the cost of traditional or cloud-native application stacks?

The answer is based on the architecture of MEC: Applications in the MEC environment can take advantage of its rich connectivity and its low latency access to the RAN to discover powerful insights that cannot be discovered in a general-purpose edge computing environment. MEC is optimized for mobile use cases for edge computing as documented in an ETSI standard. General purpose edge computing environments lack the moment-by-moment granular detail provided by the provider RAN, which manages communication for every device in the proximity of the wireless network. This is a powerful differentiator:

• Carriers can enhance their network services to dynamically maximize connection quality based on device position, behavior and context; they can dynamically assign resources to network slices; use intelligence to identify faults, deliver resources to satisfy quality of service guarantees; and ensure privacy and security.
• Customer facing MEC applications can serve both enterprises and consumers:
  • Industrial automation: MEC applications can help predict equipment failures, detect problems, optimize supply chains, manage inventory, and customize production.
  • Augmented reality (AR) and virtual reality (VR) applications help workers understand the environment around them and repair or undertake work that relies on dynamic fusion of visual and digital information.
  • Retail: MEC is required to deliver immersive in-store environments that require low latency, proximity and personalization, and offer new forms of payment.
  • Security and safety: mobile device location of one or more users can give emergency teams new tools to help in their response to emergencies.
  • Smart cities: Applications can dynamically predict future traffic load, and gauge public transit needs for citizens, tailoring each user’s experience to their own travel plans.   Information from sensors in utility infrastructure, fused with information from mobile devices, can help predict energy demands and control user environments based on personal preferences.

MEC has a powerful advantage over general purpose edge computing, namely proximity to the RAN, with low latency access to mobile devices and their status. Dynamically building a picture of each device in a connected, real-world context allows applications to respond to react in real-time to complex edge use cases.

Exploiting MEC Advantages

The juxtaposition of compute, high bandwidth connectivity, and low latency communication to endpoints makes it possible for the first time for MEC-hosted applications to analyze streaming data on-the-fly and respond before storing data. Insights can be computed immediately in-memory, in the context of a rich contextual awareness that helps to identify relationships – such as proximity, correlation, containment and more – that is required to deliver game-changing insights.

Continuous intelligence applications can

• Continuously analyze, learn, and predict, because they statefully process each update immediately, in-memory. This improves performance because saving state on disk is substantially slower.
• Analyze system behavior over time: Real-world systems evolve over time and applications need to be able to understand the joint evolution of the states of multiple devices to discover deep insights.
• Analyze in context:  Applications can discover hidden meaning in data including real-world relationships - like containment, proximity, and correlation
• Always have the answer: Applications that compute on-the-fly can analyze, estimate, and predict continuously so they always have the answer, and can react instantly.

To deliver these benefits to developers, the MEC application platform needs to be able to build and continuously update a model of the edge environment and its endpoints.

MEC is not just a “Closer Cloud”

The stateless, RESTful microservice-plus-database architecture that has been so successful in the cloud needs to be augmented with an ability to compute statefully as data flows. Accessing a database to update or gather state for analysis is a million times slower than stateful in-memory analysis at CPU speed. Locating stateful execution in the MEC and using cloud or on-prem services for storage or slow timescale analytics allows MEC to quickly deliver intelligence from floods of data that can be fused with static data by other application services. Stateful analysis of streaming data delivers continuous intelligence about devices in the real-world, from the MEC, for consumption by enterprise and cloud-hosted applications,

Although there is a trend toward faster in-memory databases, they don’t run applications and can’t find new relationships between entities. Other approaches, such as event streaming, can only act as a buffer between the real world and applications.  Reasoning about the joint meaning of a set of events requires a stateful system model that uses contextual hints, over time to build new relationships.

Swim: A Continuous Intelligence Platform for MEC

SwimOS is an Apache 2.0 licensed application platform, loosely modeled on the actor model, that builds graphs of linked actors -called Web Agents - that statefully analyze streaming data in real-time. SwimOS is delivered as a comprehensive application platform for continuous intelligence as Swim Continuum. The product includes powerful pre-built solutions for mobile providers.

Swim developers create simple object-oriented applications in Java, JavaScript or even Python. Streaming data builds a fluid, scaled-out graph of stateful, concurrent Web Agents that are analytical representations of data sources and algorithms that produce insights from them. 

Web agents concurrently and statefully analyze raw data and execute complex application logic on-the-fly, as data flows. Web Agents link to each other based on computed relationships – like proximity or correlation - dynamically building an in-memory graph.  Linked agents see each other’s state changes in memory.  They concurrently compute on their own state and that of agents they are linked to.  All state is in memory, so analysis occurs in real-time. They analyze, learn, and predict, and continuously stream insights to users, UIs, data lakes and applications. 

Swim benefits from in-memory, stateful, concurrent computation that yields orders of magnitude performance improvement over database-centric analysis because all state is in-memory and Web Agents compute immediately when data arrives. 

Examples

Smart City Traffic Prediction

Swim Continuum is used in several US cities to continuously predict traffic demands on the infrastructure. Raw data is transformed into a continuous stream of predictions and insights that are delivered to customers that need to optimally route vehicles through the city.

Swim creates a Web Agent for each sensor that reports data in the continuous stream of updates from the traffic infrastructure. Each Web Agent statefully processes raw data for its real-world sibling - a traffic sensor, light, or in-road loop. Intersection Web Agents link to their neighbors so they can see the current state of all sensors in their vicinity. They continuously predict the future state of their sensors using their own state and the states of neighboring intersections. The scaled-up application is a graph of Web Agents, each of which continuously analyzes, learns, and predicts future traffic in its vicinity. An online demo for the city of Palo Alto CA can be found here.

Network Optimization

One of the most frequently discussed features of 5G is the opportunity for carriers to deliver secure, quality of service assured network “slices”. Slicing offers both isolation and QoS, but there are challenges: Each base station needs to meet its QoS guarantees and the provider must ensure that the network can satisfy the aggregate demand. Real-time intelligence is needed to solve the capacity allocation problem. Swim is used by a major US mobile provider to manage capacity and optimize customer connection quality in real-time. A demonstration using public information sources can be found here.

Conclusion

MEC offers an opportunity to providers to deliver continuous intelligence services to their customers and to optimize their own network operations. Rather than focus on delivery of the same kinds of applications that customers can deliver from IaaS clouds, providers should take advantage of key attributes of MEC: low latency, data volume reduction, and rich interconnection to networks and services.

Most importantly MEC is uniquely situated to identify dynamic relationships between data sources, including proximity and correlation, to surface insights and respond in real time to real-world needs. Swim offers an opportunity to compute statefully on live streaming data from large numbers of mobile and static endpoints, building a dynamic graph of relationships that can immediately surface information that is impossible to see in traditional compute environments. This enables new customer-facing applications: “when a truck is near the port…” or “when my bus is approaching…” that operate at huge scale. It also enables operators to see the state of their networks in real time.