Today’s high-fidelity training systems are unable to be efficiently deployed at the point of need due to the size, weight, and power of the equipment that is required. There are limitations with how much the form factor of a system can be reduced, particularly with the slow but steady rise in power requirements on many of the CPU and GPU architectures over the last several years. A new approach is necessary to reduce the footprint of these compute resources.

In this paper we will demonstrate how we were able to reduce the footprint of our image generator systems by 50% using virtualization. This approach brings with it a unique set of challenges. If virtualized using the same approach as a typical server, challenges relating to rendering performance, hardware support, user perception, and network latency are inevitable. These challenges must be addressed to ensure the same quality of delivery with virtualization.

This paper will outline an approach to optimize virtual machines for real-time rendering and provide a testing methodology to verify that they can provide the same experience as their physical counterparts.

This approach includes:

 1.    Identification of the components for the complete virtualized solution

2.    A comparison of a physical versus virtualized IG

3.    Measuring average framerate, latency, and utilization, as well as a user’s performance completing a repeatable task within the simulation, to verify that virtualization does not have a negative impact on delivery

4.    Example use cases that both support and exclude virtualization

The methodology put forth in this paper and supported by the aggregated data points will provide the audience with an option for delivering high quality immersive training without the prohibitive footprint required by current solutions.

Keywords
EMERGING TECHNOLOGIES,ENHANCING PERFORMANCE,IMAGE GENERATOR,PORTABILITY,VIRTUALIZATION,VISUALIZATION

Additional Keywords
point-of-need, SWaP