Content area
Advances in computing and storage systems have enabled end users to run complex workloads on relatively cheap machines. These advancements have given rise to a novel infrastructure in data management: edge-cloud. Edge-cloud data management systems allow data to be stored at the edge and managed by the cloud. The availability of edge-cloud systems has led to interesting research directions in data management.
In this thesis, we deal with the problem of building data management systems that use resources from the edge and the cloud. We use the idea of the ‘cloud’ and ‘edge’ as abstractions of ‘trusted’ and ‘untrusted’ systems, respectively. This abstraction is motivated by the fact that applications deployed on the cloud (such as Gmail, Facebook, etc.) are trusted by end users. By ‘trusted’ we mean that the users trust that their data stored on the cloud is protected. This ’trust’ is based on the fact that data stored on the cloud is managed by large corporations that have a financial incentive to ensure that the data and applications they manage are secure.
In contrast with the cloud, applications or data stored on the edge (on IoT or user devices or edge datacenters) cannot be trusted by other end users unless protected by secure data storage and access protocols. Given this asymmetrical trust relationship between cloud and edge systems, any edge-cloud data management system must ensure that its design incorporates this asymmetry.
In this thesis, we examine three aspects of building an edge-cloud data management system. First, we look at minimizing coordination between edge replicas while performing read-only transactions. This is an important operation as prior research has shown that the majority of transactions performed at the edge are read operations. Thus, read-only operations at the edge must be performed efficiently, especially when reading from multiple other edge data sources. Second, we look at ensuring data integrity at the edge as well as detecting and punishing malicious edge nodes that try to modify committed data. Lastly, we look at how distributed edge-cloud transactions should be executed in a manner that takes into account the round-trip time between edge and cloud.
Our approach to these problems involves optimizing solutions for edge and cloud to ensure fast data access at the edge without intensive coordination with the cloud. We consider the cloud a trusted participant in the system. Data processing is moved to the edge, and the cloud is updated periodically. We call this approach lazy remote trust. Our work shows that this approach allows faster data processing than would be possible in traditional distributed databases. We believe this approach would be the way forward in building edge-cloud databases and applications in the future.