在行動裝置上針對3D 遊戲設計以使用者為中心之CPU-GPU
共同管理架構

在行動裝置上，圖形密集型遊戲應用程式已經越來越普及所以變得更為重要了，這類的應用程式高度需求CPU 和GPU 的資源。然而，目前的行動裝置系統的設計導致了不必要的能源浪費。這是因為缺乏考慮在應用程式所執行的階段和用戶的關注程度(使用者需求層級隔閡)，以及功率管理的設計上每個處理器只負責管理它自己的資源(處理器層級隔閡)。在本論文中，我們提出了一個以用戶為中心的CPU-GPU 架構，其目標在減少能源消耗並不顯著犧牲用戶體驗。為了彌補在使用者需求層級隔閡，我們在系統運行時將辨識用戶的需求，並相應調整各個處理器至適當的管理政策。另一方面，為了彌補在處理器層級隔閡，所提出的管理框架觀察CPU-GPU 間相互運作的情況，了解和處理器間的需求與目的。我們實現我們的框架在了三星GalaxyS4 上，並使用行動裝置上現有的3D 遊戲進行了廣泛的實驗。實驗結果表明，對於互動性強且使用者需求頻繁變換的應用程式，我們提出的架構可以減少45.1% 的能源消耗相比於目前最先進對於遊戲進行中的系統資源管理策略，並且沒有顯著犧牲應用程式中用戶的體驗。關鍵字: 異質系統, GPU, 整合排程, 資源管理, 動態電壓調節, 節能, 行動系統, 3D 手機遊戲

關鍵字

異質系統； GPU ；整合排程；資源管理；動態電壓調節；節能；行動系統； 3D手機遊戲

並列摘要

Among the mobile apps, graphics-intensive game applications, which are highly demanding of both the CPU and the GPU, exhibit their importance by owning significant popularity. However, the design of current mobile systems has led to unnecessary energy waste. This is because, first, the lack of consideration in application phases and user attention (demand-level gap) and, second, the power management design for each processor administers in its own way (processor-level gap). In this paper, we propose a user-centric CPU-GPU governing framework which aims at reducing energy consumption without sacrificing user experience significantly. To bridge the gap at the demand level, we identify the user demand at runtime and determine accordingly the appropriate governing policies for respective processors. On the other hand, to bridge the gap at the processor level, the proposed framework interprets the frequency scaling intents of processors based on the observation of the CPU-GPU interaction and the processor status. We have implemented our framework on Samsung Galaxy S4, and conducted extensive experiments with real-world 3D gaming apps. The experiment results showed that, for an application being highly interactive and frequent phase changing, our framework can reduce 45.1% energy consumption when compared with the stateof- the-art policy without sacrificing user experience significantly. Keyword: Heterogeneous system, GPU, joint scheduling, resource management,DVFS, energy efficient, mobile system, 3D mobile games

並列關鍵字

Heterogeneous system ； GPU ； joint scheduling ； resource management ； DVFS ； energy efficient ； mobile system ； 3D mobile games

參考文獻

[2] Bren Mochocki, Kanishka Lahiri, and Srihari Cadambi. Power Analysis of Mobile 3D Graphics. In Proc. of IEEE DATE, pages 502–507, 2006.

[6] Frances Yao, Alan Demers, and Scott Shenker. A Scheduling Model for Reduced CPU Energy. In Proc. of IEEE FOCS, pages 374–382, 1995.

[8] Beilei Sun, Xi Li, Jiachen Song, Zhinan Cheng, Yuan Xu, and Xuehai Zhou. Texture-Directed Mobile GPU Power Management for Closed-Source Games. In Proc. of IEEE HPCC, pages 348–354, 2014.

[9] Yan Gu and S. Chakraborty. Power Management of Interactive 3D Games Using

[10] Yan Gu and S. Chakraborty. Control Theory-based DVS for Interactive 3D Games. In Proc. of IEEE/ACM DAC, pages 740–745, 2008.

國際替代計量

在行動裝置上針對3D 遊戲設計以使用者為中心之CPU-GPU 共同管理架構

全文下載

主題瀏覽