Disconnected, denied, intermittent, or low-bandwidth – Getting Started with Edge Computing on AWS

While the specific term Disconnected, Denied, Intermittent, or Low-bandwidth (DDIL) emerged from the US Department of Defense, it captures well the circumstances faced by a group of use cases seen across industries. It refers to edge computing in situations where network connectivity is unreliable, constrained, or completely unavailable. In such scenarios, traditional cloud-based computing approaches might not be feasible, and edge computing can play a crucial role in enabling data processing and decision-making at the source. Services such as AWS Snowball, AWS Snowcone, and AWS IoT Greengrass can help in these cases.

Disconnected

In environments where there is no network connectivity at all, devices must be able to operate independently, processing data and making decisions locally. This typically requires the implementation of efficient data storage, processing capabilities, and pre-trained ML models to allow the device to function effectively without access to the cloud. These are usually the same situations where standard data center environmental controls such as HVAC, particulate filtration, and physical security measures are lacking.

As an example, an energy company is conducting a survey for oil in a remote desert location. They are driving a special kind of truck, known as a thumper, that shakes the ground with hydraulically driven metal pads. The shock waves travel through the earth to instruments known as geophones. The data thus collected must be analyzed to produce a three-dimensional map of what is going on underground. Local ruggedized computers can be used for this first step, but the second step requires a highly trained and experienced geologist to make a judgment call about what is seen in this map – in a similar way that a radiologist is needed to properly analyze an x-ray image.

While they may be able to make voice calls or transmit data via satellite connectivity, this is both expensive and very low throughput – measured in kilobits per second. Wouldn’t it be great if those determinations – inferences, in ML parlance – could be made on the spot? It would be even better if every time such a team came back into a place with better connectivity, the data they collected automatically synchronized with the cloud. This would constantly improve the experience of the system. In AI parlance, this is known as training. Such a model would necessarily be quite large, and the AWS cloud is the perfect place for this training to happen.

While most AWS services are dependent upon consistent high-speed connectivity to the AWS control plane, AWS Snow Family is different. These devices maintain their own local control- and management-plane services yet expose the same APIs and management constructs as their in-region equivalents.