Scalable and Fault-Tolerant Storage and File System Services with Non-Blocking Synchronization for Private Clouds

Abstract

We present two system services – the storage service and the file system service designed for private cloud environments to facilitate file sharing across different virtual machines (VMs). Our services are scalable, fault-tolerant, and deliver excellent performance. These system servers are implemented as unikernels running atop of the Xen hypervisor. Additionally, our storage service can leverage NetBSD code, enabling support for a wide range of both legacy and modern storage devices, such as NVMe. Furthermore, the storage service addresses the challenge of transparent fault recovery for storage, a complex task for stateful subsystems, without incurring significant overhead – a well-known challenge in storage systems. Our file system service is designed to be copy-free, enhancing overall performance. We have also designed an inter-VM communication (IVMC) mechanism that fosters scalability and reliability by leveraging lock-free concurrent ring buffers. Since this mechanism is lock-free, our system services communicate with application VMs in a more scalable manner compared to traditional ring buffers used in hypervisors such as Xen. Our lock-free design also aids in restoring storage states during the fault recovery process of the storage server. Our evaluation results demonstrate that our system services achieve performance comparable to that of Linux.