We research next generation Cloud infra-software, Stella Project, which includes the two core development technologies for multiple SLA guarantee of multiple users. First core development is to integrate the scheduling of the CPU, network, and storage resources. Second is Cloud-wide SLA guarantees including resource admission control, job allocation, and management techniques.
Our research group builds a next generation network system. Recently, the paradigm of network systems is shifting from the closed system to the open system. Based on the drastic evolution, we carry out research on two principal network systems: 1) SDN-based network virtualization, and 2) high-performance kernel-based networking.
Software-defined network (SDN) is a novel paradigm of network architecture which eases network management and advances network evolution. We develop a network hypervisor that creates multiple virtual networks on a single physical network based on the SDN concept. SDN-based network virtualization provides perfect isolation and programmability of individual virtual networks, so that it is regarded as a critical technology for realizing artificial intelligence or IoT services.
The operating system such as Linux supports the general-purpose network processing, but its implementation incurs inevitably considerable overheads for high-speed packet processing. To avoid such overheads, kernel bypassing techniques such as DPDK and RDMA have been proposed, but they require significant modifications to user applications. Our goal is to leave applications unmodified but achieve the high performance. We develop high-performance kernel-based networking by applying new optimization schemes as follows.
In general, the real-time task scheduling does not work in a virtual machine-based system because the physical resource scheduling and management is done at the lower layer, namely, the virtualization layer. And recent system virtualization approaches are focusing on performance issues; however, Real-time(RT) systems are affected by the underlying hardware and its configuration. Depends on the configuration, system parameters such as execution time of tasks, execution period, system utilization, etc are affected. RT system schedulers have a unique scheduling policy that all the real-time tasks can finish their jobs within a given deadline always. But, the virtualization layer is focusing on the performance issues. Therefore, previous Virtualization techniques is not able to achive real-time process. Our research interests are as follow.
Copyright © 2019 | Operating Systems Lab.