Please note: This PhD defence will take place in DC TBD.
Kenneth Ryan Hancock, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Ali José Mashtizadeh
Storage devices are orders of magnitude faster than a decade ago. Furthermore, due to increasing numbers of high bandwidth PCIe slots on newer servers, storage IO bandwidth has started to match total memory bandwidth. This critical break point leads to a need to rethink current storage solutions, and relook at older idea’s such as the Single Level Store (SLS) operating system. In an SLS operating system, the OS takes the responsibility of persistence from the user and users write programs as if they never crash. Single level stores merge the virtual memory, and storage subsystems of the operating system and whose concepts can be applied to areas such as databases, serverless computing, and file systems.
This thesis outlines ObjSnap, a storage system designed and built for two modern single level store system’s, Aurora, and MemSnap. ObjSnap is a transactional, copy-on-write object store that is able to outperform other storage systems by up to an order of magnitude. Further, my research extends into fields in which SLSes have promising applications, such as serverless. Through this work, I created a benchmark used to evaluate serverless orchestrators, called OrcBench. Further, I introduce, SlimSys, a framework which tackles security issues found within containers, the preferred deployment mechanism used throughout serverless.