“Smart edge” is Booz Allen’s term for an emerging trend in the internet of things — the shift to imbue sensors and devices in a contested environment with the ability to make smart, autonomous and coordinated decisions in a low-bandwidth or on an unreliable network. Smart edge operations depend upon data local to devices and from other coordinating devices and sensors in close proximity, and can work independently with no connection to the internet. Independently operating computing resources, such as cloudlets, are important enablers of the smart edge paradigm. A cloudlet is a mobility-enhanced small-scale data center that is located at the edge of the enterprise and is available for use by nearby devices. In this blog, we will cover cloudlets and their relevance as a smart edge enabler.
The half-life of IT has compressed, accelerating many technical evolutions. The abstraction evolution came in as we transitioned from mainframes to client server, and multitiered systems to web-based systems where infrastructure, business logic and user interface are decoupled. The commoditization of services and infrastructure has enabled the creation of the cloud, mobile computing and the internet of things, which has allowed for processing at the edge and the ability to do distributed computing. This will in turn enable the next wave of evolution — “chip as a platform” — which will unleash an era of explosive growth in the IT industry as it moves more and more to the edge. “Chip as platform” will bring both benefits and challenges. For example, in the DoD it will mean:
- Increasingly sophisticated sensors on UAVs and other platforms. This will drive the need for intelligent ISR and processing at the edge to short circuit the FPED cycle.
- Greater attack surface. The rapidly increasing number of sensors will continue to drive the importance of cyberdefense and resilience.
- Exponential collection of raw data. This will increase the need to shorten the cycle from “data to action.”
Edge computing will also be disruptive as we move from “dumb sensors” to connected sensors that are smart, self-aware and ubiquitous. Different integration and processing platforms will be required to enable the smart edge. Hub-and-spoke models allow devices to communicate with a tactically deployed central node, while mesh networks are based on a decentralized model where all devices are peers. Cloudlets support both of these models and can be key enablers of IoT and edge computing.
What is a cloudlet and how does it enable smart edge?
A cloudlet is a trusted, resource-rich computer or cluster of computers that may or may not be connected to the internet and available for use by nearby devices. A cloudlet supports resource-intensive and interactive applications and provides powerful computing resources to devices with lower latency. The term cloudlets originated in the mobile-edge computing industry initiative created by the European Telecommunications Standards Institute.
There are two main architectural approaches to cloudlets. The first is the transient cloudlet (Figure 1), based on a standard hub-and-spoke model, where mobile users access a nearby cloudlet over a wireless LAN/RAN. The transient cloudlet relies on a resource-rich computer infrastructure, providing data storage and computing service accessible to mobile devices through wireless networks, mainly cellular and WLAN.
The second type is the mobile cloudlet, where a set of resource-rich mobile device devices, referred to as cloudlet nodes, can connect to each other on a mesh network and provide and consume services. The mobile cloudlet relies on peer-to-peer mesh communication, whereby a group of nearby mobile devices can connect via secured Wi-Fi or Bluetooth. In this model, each mobile device shares computing service as nodes on the mesh, leveraging distributed computing principles.
Challenges and lessons learned
As part of our research into cloudlets, we encountered a number of challenges that each cloudlet implementation approach faces.
Transient cloudlet: This implementation model faces three main challenges, including: rapid (agile) provisioning to reduce delay and address the user mobility; VM handoff, to seamlessly migrate the offloaded services on one cloudlet to the next; and cloudlet discovery to enable distributed mobile devices to discover, select and associate with the appropriate cloudlet among multiple candidates before it starts provisioning.
Mobile cloudlet: Today, each virtualized system gets its own set of resources allocated to it and does minimal sharing. The majority of the resources required for VMs is taken up by the hypervisor and each guest OS, requiring a much larger footprint.
In addition, transient cloudlet implementations struggle with cluster management (distributed mesh), scaling, desired state reconciliation, multihost networking, service discovery, load balancing, security and rolling updates. Container technologies, such as Docker Swarm, provide many of these capabilities out of the box, as well as isolation of VMs, and share resources as reusable images (OS, database, application services, etc.), allowing them to be more efficient, faster and more lightweight. This is done in a distributed manner, where the platform enables “self-discovery” through a mesh network (Swarm). The container wraps a piece of software in a complete filesystem that contains everything needed to run: code, runtime, system tools and system libraries. They start instantly and use less RAM. Images are constructed from layered filesystems and share common files, making disk usage and image downloads much more efficient. Some of the challenges this approach still faces includes cloudlet node discovery. Multicast DNS or Wi-Fi P2P can be used for remote provisioning and a secure REST service could be deployed on each cloudlet manager node.
Cloudlets are a key element to be considered as part of an IoT smart edge computing strategy, especially where the solution needs to provide:
- Low end-to-end application latency (real-time)
- Maximum transaction rate between device and local “cloudlet” for optimal compute results (interactive)
- Local communications to private networks for performance, privacy and security (secure)
- Real-time insights from data at the point of capture, minimum cloud ingress bandwidth (analytical)
- Rapid introduction of network and other functions in a radio area network (RAN) with dynamic filtering rules (distributed)
Careful analysis of these types of requirements will determine the best architectural approach (transient versus mobile cloudlet) to support smart edge solutions.
This article was co-written by Ki Lee, principal at Booz Allen Hamilton.
All IoT Agenda network contributors are responsible for the content and accuracy of their posts. Opinions are of the writers and do not necessarily convey the thoughts of IoT Agenda.