NSF Award
2324923
This project aims to develop a high-performance cloud operating system (OS), called Octopus OS,<br/>that can adapt to the dynamic cloud environment. Instead of running on a physical machine,<br/>the OS now is often running in a virtual machine (VM) in the modern cloud where the resources<br/>are highly dynamic due to resource sharing and VM migration. However, the traditional OS<br/>cannot manage the virtualized resources efficiently due to being unaware of the resource dynamics,<br/>leading to low resource utilization and poor application performance in the cloud. Therefore,<br/>the proposed Octopus OS would create and manage the accurate abstractions of the dynamic<br/>resources, and develop new algorithms to fully leverage the unique characteristics of the virtualized<br/>resource, which would greatly improve application performance and resource utilization in the<br/>dynamic cloud. This work can advance the cloud OS by serving as a framework for optimizing the<br/>utilization of various virtualized resources. The outcomes will be released for public use. Hands on<br/>projects will be created during the project development to enrich existing courses related to<br/>OS and cloud computing. This project will involve students from underrepresented groups and promote<br/>research in computer systems for all undergraduates.<br/><br/>Four steps will be taken to achieve the project goal. First, the mismatches in major virtualized<br/>resource abstractions (i.e., CPU, memory, and cache) will be analyzed in a controlled environment<br/>to understand their impacts, including resultant system abnormal behavior and application<br/>performance degradation. Second, accurate resource abstractions will be created to fix the identified<br/>mismatches using probing techniques. Specifically, a set of micro-benchmarks, called vProbers,<br/>will be developed to profile the dynamic nature of the virtualized resources without relying<br/>on the hypervisor support or application porting, making this solution practical in the emerging<br/>multi-cloud environment. The accurate resource abstractions probed within Octopus OS can also<br/>be leveraged by cloud application and language runtime, and are critical to provide resource visibility<br/>for cross-cloud optimizations. Third, the new abstractions will be exposed to Octopus OS to<br/>make it virtualization-aware. New algorithms will be developed to fix existing system abnormal<br/>behavior and unlock the potential performance benefits of the dynamic resources. Fourth, experiments <br/>with resource-demanding benchmarks in various public clouds will be conducted to evaluate the ability of <br/>the proposed Octopus OS to adapt to the dynamic resources in the multi-cloud environment.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
