# A Benchmarking Framework for Interactive 3D Applications in the Cloud

**Authors:** Tianyi Liu, Sen He, Sunzhou Huang, Danny Tsang, Lingjia Tang, Jason Mars, Wei Wang  
**arXiv:** [2006.13378](https://arxiv.org/abs/2006.13378)  
**Date:** 2020-06-23  
**Summary page:** https://www.summarizepaper.com/en/paper/2006.13378v2/

## Key points

- The paper addresses the significance of interactive 3D applications in cloud computing.
- Lack of public research on designing efficient cloud systems for supporting these workloads.
- Introduction of Pictor, a novel cloud graphics rendering research infrastructure leveraging AI.
- Development of a benchmark suite within Pictor comprising six interactive 3D applications for performance analysis.
- Uncovering previously unknown performance bottlenecks in cloud-based 3D graphics rendering systems.
- Implementation of optimizations based on findings leading to a significant average improvement in frame rate.
- Demonstrating the effectiveness of Pictor for enhancing efficiency and performance of cloud-based interactive 3D applications.

## Summary

The paper "A Benchmarking Framework for Interactive 3D Applications in the Cloud" by Tianyi Liu et al. addresses the growing significance of interactive 3D applications in cloud computing. These applications have become increasingly popular with the rise of cloud gaming and virtual reality technologies. However, there is a lack of public research on designing efficient cloud systems to support these workloads. To bridge this gap, the authors introduce Pictor, a novel cloud graphics rendering research infrastructure that leverages artificial intelligence (AI) to simulate human interactions with complex 3D applications. Pictor also enables detailed performance measurements for the intricate software and hardware stack utilized in cloud-based 3D graphics rendering. The researchers developed a benchmark suite within Pictor comprising six interactive 3D applications to analyze performance characteristics and identify new bottlenecks in cloud-hosted 3D applications. Through performance analyses using these benchmarks, they were able to uncover previously unknown performance bottlenecks in state-of-the-art cloud 3D graphics rendering systems. Subsequently, two optimizations were implemented based on their findings, resulting in a significant average improvement of 57.7% in frame rate. This demonstrates the effectiveness of Pictor as a tool for enhancing the efficiency and performance of cloud-based interactive 3D applications. Overall, this study contributes valuable insights into optimizing cloud systems for supporting interactive 3D applications and highlights the importance of developing robust research infrastructures such as Pictor to drive advancements in this domain.