Computer data centers today are experiencing a disruptive trend by virtue of becoming software-defined and programmable. Software-defined networking (SDN) and software-defined storage (SDS) have been widely deployed in today's data centers to provide multi-tenant network and storage services. However, due to the lack of coordination between the network and storage software, there are still performance, scalability, and reliability challenges in data centers. As the data center is evolving with software-defined functions (hitherto done through hardware devices), it is the right time to fundamentally address these classical challenges. This project leverages a cross-layer research approach to conduct the fundamental systems and architecture study on the software-defined network and storage. The proposed research activities will benefit the infrastructure and platform development for large-scale data centers. The project will involve active industrial collaborations to carve a path for technology transition to practice. Furthermore, the project will integrate the research developments into new courses that center around network and storage techniques. This project will also recruit and train underrepresented and high school students through various research programs and summer camps.
This project investigates a new software-defined architecture that co-designs the system stacks of both SDN and SDS while refining the functionalities of each component. If successful, the project will fundamentally advance design principles and techniques of software-defined data centers. First, the project proposes to refine the functionalities of the control and data plane in SDN and SDS. Such a new architecture enables state sharing between SDN and SDF, which facilitates the global resource management and I/O scheduling to address the performance challenge. Second, to overcome scalability challenges, this project rethinks the design and implementation of concurrency control in the storage stack and proposes an in-network lock manager to scale the concurrent storage accesses. The project further exploits the hardware parallelism available in multi-core CPU chips to speed up the network packet processing. Third, the project analyzes the limitation of existing data replication protocols and proposes an in-network conflict detection scheme to achieve both linear scalability and strong data consistency for replicated storage. Finally, the project applies the proposed techniques to representative multi-tenant storage services such as distributed databases and key-value stores, and new distributed applications.
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.
