Elevating the Edge to be a Peer of the Cloud

Abstract: The surge of cyber-physical systems around us has led to the emergence of novel use-cases like smart surveillance and smart cars. These use-cases either involve high amounts of machine-to-machine communication or are highly interactive, making low-latency response to events critically important. Traditional cloud computing paradigm is unsuitable for such applications due to high network latency to datacenters, leading to the emergence of the edge computing paradigm. Edge computing allows application logic to be hosted near the edge of the network, allowing low-latency responses.

In an edge-computing ecosystem, application components work together (both within a single application and across different applications sharing the edge infrastructure), sharing information with each other through a data management platform. Contemporary edge computing solutions like Azure IoT Edge are heavily dependent on cloud-based backends for data management. This design of a cloud-dependent edge does not work well for geo-distributed applications. We intend to make edge computing a peer of cloud computing, such that it offers similar properties like self-manageability, fault-tolerance and the illusion of infinite resources. This would make the task of edge-based application development easy as is the case with cloud applications. The objective is to build a data management platform that is based on the edge-cloud infrastructure continuum.

Such an edge-based solution to data management has many research challenges. Firstly, data placement policies of off-the-shelf platforms like Cassandra or Kafka don't take into account the geographical heterogeneity in edge infrastructure. Edge infrastructure is more vulnerable to geographically correlated failures, necessitating wide-area data replication. Replication comes at the cost of high-overhead consistent reads/writes which is detrimental to application performance. Furthermore, skews in data workloads lead to hotspots on edge resources, which are already scarce.  The poster presents these challenges and our ideas on solving them.

Explanation of Demonstration: This demo intends to showcase a smart camera surveillance with its vehicle tracking application. With smart cameras' capability of computing and network communication, we are able to bring down the majority of processing at the edge of the network. Relative research has been published at STTR: a system for tracking all vehicles all the time at the edge of the network.