動態調整快取預取應用於影像壓縮標準

身處在21世紀的時代之下，我們每天的生活中缺少不了影像、視訊等，使這些multimedia資訊可以廣泛地在世界流動，compression standard功不可沒，然而當播放這些影像、視訊的時候，會需要系統架構能有效地支援才不會造成影像播放的延遲。 cache memory身為系統中協調的角色，雖然可以緩衝CPU和main memory間速度差異所帶來延遲時間，但仍然會有像是CPU在cache中找不到所需資料的情況，因此發生cache miss，因為需要到main memory中存取資料，需要花費較多時間而造成系統效能下降；為了可以有效改善cache miss的發生頻率，許多學者對cache memory架構做出了創新的研究，然而現今的系統整體架構發展已大致趨近飽和，在不改動現有基本memory架構之下，出現cache prefetching方法，可以有效改善cache miss發生。為了改善compession standard在系統所發生的延遲，本篇論文使用multimedia prefetching方法中的neighbor prefetching，將其存取3*3矩陣鄰域擴展成5*5及7*7的矩陣鄰域，再搭配動態調整cache prefetching的方式來改善compression standard的cache miss rate 。在本篇論文中，我們以動態方式調整cache prefetching degree，然後在6支compression standard平均的L2 cache miss rate上，相較於傳統的prefetching方法的OBL及correlation-based分別得到63.73%及79.67%的改善，而對於multimedia prefetching方法中3*3 neighbor則是得到了37.80%的miss rate改善。

關鍵字

快取預取；壓縮標準

並列摘要

Living in the 21th century of intensive telecommunication, we can’t live without images and videos. The multimedia information can be able to be broadcast extensively, compression standards play an important role. In order to prevent the delay of displaying videos, compression standards need the effective system architecture to effective support of the system architecture when we display these images and videos. As a coordinator in the system, cache memory is capable to buffer the latency caused by the speed gap between CPU and main memory. However, when CPU can’t find require data in the cache, then cache miss happens. In order to improve the frequency of cache miss, many scholars proposed novel research on cache memory architecture. Nevertheless, the development of cache memory is almost at the limit. Based on not modifying the basic memory architecture, the cache prefetching approach came up. This technique can improve occurrence of cache miss. In order to improve the delay caused by the compression standard in the system, this thesis adopts neighbor approach of multimedia prefetching. We expand the access range of 2 d neighbor block from 3*3 matrix to 5*5 and 7*7 matrix. Then we use the proposed approach with dynamically configured cache prefetching to improve cache miss rate of the compression standard. In the thesis, we configure the cache prefetching degree dynamically. We have 63.73% and 79.67% improvement in the average level 2 cache miss rate over the traditional prefetching approach OBL and neighbor. Compared with the multimedia approach, 3*3 neighbor prefetching, we have 37.80% improvement.

並列關鍵字

Cache Prefetching ； Compression Standard

參考文獻

[1] Daniel F. Zucker, Michael J. Flynn and Ruby B. Lee ,“A Comparison of Hardware Prefetching Techniques For Multimedia Benchmarks ,” Proc. of the Third IEEE International Conference on Multimedia Computing and Systems, pp. 236 -244, June 1996.

[2] Zhou Ni, Qiao Fei and Yang Huazhong, “A Hybrid Cache Architecture with 2D-based Prefetching Scheme for Image and Video Processing,” Proc. of the 2013 International Conference on Communications and Signal Processing (ICCSP),

pp. 1092-1096, April 2013.

[3] Z. Xu, S. Sohoni, Min Rui and Y. Hu, “An Analysis of Cache Performance of Multimedia Applications,” IEEE Transactions on Computers, Vol. 53, Issue: 1,

pp. 20-38, Jan 2004.

國際替代計量

動態調整快取預取應用於影像壓縮標準

未授權

主題瀏覽