Linear Model Analysis of Real-Time Rendering - Real-Time Rendering: Computer Graphics with Control Engineering - page 20

In this research, we focus on exploiting a current trend in hardware technol-
ogy that provides fine resolution in geometry control, known as tessellation . Since
geometry is the primitive construct of any object in 3D space, it becomes a natural
choice as one of the modelling variables in our framework. In brief, tessellation is
the process of sub-dividing surfaces into smaller shapes with the objective of gen-
erating higher resolution information of the 3D model. Tessellation, also known as
a subdivision technique, is a well researched field in computer graphics and had
been adopted widely in many interactive rendering applications because of the visual
acuity it provides. However, only recently has graphics hardware provided sufficient
support for tessellation-based techniques in applications [30].
We introduce our modelling framework via experiments in two interactive render-
ing applications that use subdivision techniques in rendering load control. We aim to
establish the fundamental validity of a system-based approach to modelling the ren-
dering processes in applications similar to those selected in these experiments. Since
rendering tasks are inherently complex in real-world applications, we provide a sys-
tematic extension from a single-input-single-output (SISO) model of the rendering
process to a multiple-input-single-output (MISO) model that more closely resembles
a broader class of applications. We hope that this progression along with the system
modelling principles and fundamental considerations related to 3D rendering will
enable readers to appreciate the value of this framework and acquire the necessary
knowledge for its implementation.
Current research in rendering workload characterisation [16,17] and rendering
time estimation [18,19] strives to profile the attributes of rendering without providing
a systematic way to control the process. Often, the user is expected to arrive at some
form of a primitive control strategy based on profile information. This often requires
several attempts to re-evaluate control strategy and ad hoc refinement steps are often
needed to remove major rendering bottlenecks.
This motivated us to attempt to utilise a systems perspective to model the render-
ing process. In this chapter, we demonstrate that accurate models can be obtained
via our data-driven framework and extend this framework by introducing the use of
a controller that can track and regulate frame rates with guaranteed performance.
In comparison with other work, our research offers the following benefits:
• Our framework does not require the pre-processing of the 3D content
utilised in other research [20,21,22]. Its performance is not limited to static
pre-processed geometry and scenes.
• Our approach leverages hardware-accelerated technology (tessellation) that
provides smooth transitions in geometry scaling unlike techniques that may
generate visual hysteresis [21,22,23].
• The outputs of the derived rendering models exhibit very high accuracy
when compared to actual rendering process outputs.
• When the derived rendering model is used in conjunction with a suit-
able controller, the closed-loop system can produce guaranteed frame
rates. The self-correction process occurs entirely online during runtimes
unlike current techniques that may require repetitive and labour-intensive
offline evaluation.