Virtual Open Systems Scientific Publications

Introduction

In recent years, the widespread adoption of virtualization technologies has led to a growing demand for effective and efficient management of hardware resources in data centers, cloud computing, and high-performance computing environments. The management of devices in these environments can be challenging, especially when dealing with multiple VMs and frequent reconfigurations. To accelerate complex workloads, Field Programmable Gate Arrays (FPGAs) are more and more used to provide a fast-deployment, fast-prototyping solution to custom problems, with an increasing interest in Artificial Intelligence (AI) and Machine Learning (ML) domains. Virtualization of FPGA resources also enables scenarios where a single FPGA can provide multiple devices to different VMs, to enhance the workload and security.

Addressing some of the VMs' guest acceleration is already possible by using the Single Root I/O Virtualization (SR-IOV) technology to allow multiple VMs to share the same physical device providing direct access to the hardware with minimal to no overhead. However, SR-IOV usage is not always straightforward and can lead to complex scenarios, especially when there is the need to change the VF configuration on the fly which requires the removal of every VF from each VM. This is not a problem in applications with fixed and predictable workloads or with dedicated resources for the VMs, but in environments such as cloud computing, software as a service, high-performance computing, and edge computing where workloads can vary rapidly, resources need to be shared among different VMs, and devices need to be reconfigured and reassigned, the limitations of SR-IOV become more evident. In these cases, a solution that allows automatic assignment and reconfiguration of virtual functions, can significantly simplify device management and improve system efficiency.