Coordinated Resource Management for Cloud-based Specialized Operating Systems
Containerization has recently gained significant interest among cloud providers and users due to its ease of deployment and lightweight virtualization capabilities. The key feature of these approaches is the sharing of a single Linux OS instance among each active container environment. By sharing a single OS environment containers are able to exhibit much greater resource efficiency by avoiding the replicated state that arises in heavy-weight virtualization. However, the container approach has the drawback of relying on a single general purpose OS, whose design decisions might not align with those of the application. A recent resurgence of Specialized Operating Systems and Runtimes (SOS/Rs) has shown that significant performance and efficiency improvements can be made by codesigning the OS and application runtime environment together. The goal of this project is to enable the flexible deployment of these SOS/Rs alongside current container based solutions. This work will allow application services to be deployed on an optimal system software stack, and to react dynamically to changing hardware resource availability.
Collaborative effort with the Prognostic Lab at The University of Pittsburgh.
- Towards a Practical Ecosystem of Specialized OS Kernels
- ConCORD: Easily Exploiting Memory Content Redundancy Through the Content-aware Service Command
- CSR: Small: Collaborative Research: Flexible Resource Management and Coordination Schemes for Lightweight, Rapidly Deployable OS/Rs
- Nautilus Aerokernel
- Multiverse: Easy Conversion of Runtime Systems Into OS Kernels via Automatic Hybridization