Cloud XR Streaming: Virtualization Enhancing Resource Utilization and System Flexibility

Virtualization Technology creates independent operating environments by abstracting computing resources. It provides a virtualized computing environment for applications through software simulation and hardware resource abstraction and isolation, achieving resource isolation, sharing, and dynamic allocation. This technology significantly improves resource utilization and system flexibility. Below are common Virtualization Technologies used in Real-Time cloud XR Streaming:

GPU Virtualization

Passthrough Mode: This mode directly assigns a physical GPU to a virtual machine, with almost no performance loss and no need to modify hardware drivers. It is widely used by many public cloud service providers. GPUs supporting passthrough can fully utilize their performance without quantity limitations and without affecting GPU functionality.

GPU Slicing Virtualization: This method divides GPU resources from two perspectives: time slices and memory slices. Time slices, similar to CPU process scheduling, share a physical GPU’s computing resources among multiple vGPUs, with each time slice typically between 1-10ms. Memory slices allocate GPU memory to each vGPU, such as dividing a 16GB GPU into 16 vGPUs, each with 1GB of memory. Although there is a slight performance loss, this method offers greater flexibility and supports hot migration.
GPU SRIOV Virtualization: SRIOV (Single Root I/O Virtualization) technology splits a PCI resource (PF) into multiple small devices (VF) that comply with PCI specifications. This method has lower flexibility, usually determined by hardware how many vGPUs a physical GPU can be split into, and cannot be further expanded.

Graphics API Virtualization

Virtualization at the graphics API layer, such as OpenGL and DirectX virtualization.

Operating System Virtualization

Technologies like KVM, VMware, vSphere, and VMware Workstation.

Multi-Process Virtualization

This virtualization achieves process-level isolation within a single operating system, allowing multiple rendering processes to run simultaneously on one server, improving CPU and GPU resource utilization

Advantages of Virtualization Technology

Compared to traditional physical resource allocation methods, virtualization technology offers the following advantages:

Resource Utilization Improvement: Virtualizing physical resources into multiple logical resources improves overall resource utilization.
Flexibility Enhancement: Achieves resource isolation, sharing, and dynamic allocation, enhancing system flexibility and scalability.
Security Improvement: Prevents resource conflicts and potential security vulnerabilities through resource isolation and security control.
Reliability Enhancement: Improves system reliability and fault tolerance through resource redundancy and backup.

LarkXR: Real-Time Cloud XR Streaming Solution

The underlying architecture of Paraverse’s Real-Time Cloud XR Streaming Solution, LarkXR, provides GPU resource pooling management capabilities. It dynamically allocates GPU usage by intercepting graphical processing interfaces and isolates processes for upper-layer applications, providing a containerized environment for independent operation. This supports a shared mode of one machine with multiple cards and one card with multiple applications on high-density GPUs, achieving ultra-fine-grained resource allocation and improving resource utilization.

In scenarios of short-term high concurrency, cloud providers in the same region and with the same card type may find it challenging to meet large-scale application demands. LarkXR’s underlying heterogeneous and upper-layer homogeneous features allocate different numbers of user requests based on different card types, offering more choices for large-scale applications.