Load Balancing for Hybrid NoSQL Database Management Systems

In the field of NoSQL database systems nowadays, different systems have to make choices on whether to optimize for consistency, availability, and partition tolerance. Some of them has limitations on the number of concurrent served clients. Others might require the memory to be as large as the data in disk. Different design principles would lead to different latencies under different workloads and queries per second (QPS). To benefit from multiple design principles, some companies deploy hybrid databases for their applications. However, the application developers have to predefine schemes which assign tables to databases. Frequently used tables stay in more responsive nodes, while others stay in more reliable systems for permanent storage. When some parts of less-frequently accessed tables get unexpected amount of access accidently, this kind of systems cannot accomodate the changes and migrate data to more responsive nodes.
In this thesis, we propose a load balancer with capabilites to dynamically detect hot spot node and migrate data to nodes with spare throughput capabilites. We also provide an extendable/flexible database interface to attach to different databases, and expect maintainers to directly do CRUD or other advanced operations without additional tuning on the schema.