SkyhookDM: Data Processing in Ceph with Programmable Storage

With ever larger data sets and cloud-based storage systems, it becomes increasingly more attractive to move computation to data, a common principle in big data systems. Historically, data management systems have pushed computation nearest to the data in order to reduce data moving through query execution pipelines. Computational storage approaches address the problem of both data reduction nearest the source as well as offloading some processing to the storage layer.