rShare: Alleviating long startup on the Cloud-rendering platform through de-systemization
Dongjie Tang,Marc Mao,Yong Yao,Cathy Bao,Qiming Shi,Chao Xie,Randy Xu,Mohammad R. Haghighat,Yun Wang,Zhengwei Qi,Haibing Guan,Xiaojie Cao
DOI: https://doi.org/10.1016/j.sysarc.2023.103009
IF: 5.836
2023-01-01
Journal of Systems Architecture
Abstract:WebGL paves the way for the great development of lightweight rendering applications (e.g., web games). However, the target audience of Cloud-rendering frameworks (e.g., Google Stadia and gRemote) mainly focuses on native applications, called traditional Cloud-rendering frameworks. Even though traditional frameworks cover all the rendering functionalities and both light-and heavy-weight applications can run on them, such frameworks face long startup latency, imposing a significant challenge on lightweight applications. To reduce the startup latency, many researchers render objects locally, but the system-level redundancy still affects the performance of lightweight rendering applications negatively (especially in the application-initialized stage).In this paper, we propose rShare, a new rendering framework dedicated to lightweight applications. To reduce the startup latency of lightweight applications, rShare desystemizes the entire user space into the transmitted protocol. To ensure the portability of the entire system, rShare establishes two vHALs (virtual Hardware Abstraction Layer) (i.e., peripheral vHAL and rendering vHAL) within the protocol. To boost the CPU running efficiency at the initialized stage, rShare establishes an optimized mechanism called flexTLB, allowing various page sizes to work together. The experiments show that rShare can improve the CPU executing efficiency by 20%similar to 50% and reduce 90% startup latency compared to the CARE, a traditional Cloud-rendering framework proposed by Intel.