基於軟體定義監控應用之多媒體網路晶片

Software-Defined Monitoring for Multimedia in Network-on-Chip

李宣儒(Hsuan-Ju Li)

指導教授 ： 熊博安

國立中正大學/工學院/資訊工程學系/碩士(2015年)

摘要

隨著科技的進步，網際網路服務成為生活中的必需，例如多媒體服務，尤其行動載具的發展迅速，帶來了高容量多媒體傳輸需求。多媒體串流技術，是用來解決以往影音檔需下載完成後才可播放的問題，當傳輸速率高於播放速率時，可以達成毫無延遲的播放影片。但也因此被一些外在環境因素所限制，例如網路頻寬、裝置上處理器的計算能力以及記憶體大小。因此針對串流多媒體，需要一個彈性化的機制，使得它有能力根據外在環境因素，即時性的調整傳輸品質。
本論文提出一套在晶片網路中實現軟體定義機制(SDM)，進行快速分析網路流量。SDM於多媒體串流封包接收中，在大小為5的GOP，根據1%的封包丟棄頻率以及25.5 dB以上之PSNR表現水準，做到理想的影片品質丟棄封包方式，同時也能符合行動載具在無線網路的環境。
SDM架構包含三種元件解析器、規則表以及規則執行器，實作於晶片網路的系統中，並且與傳統的串流軟體結合在SDM設計中，藉由晶片網路通訊架構可平行化處理多封包流，進而提升多媒體串流的效能。實驗結果指出，SDM架構雖然在影片框架偵測上時間不及於單純軟體，但是在處理多媒體串流的處理上，SDM能夠為CPU帶來最多13倍的效能改善。

關鍵字

晶片網路 ； 軟體定義監控 ； 網路流

並列摘要

Internet services have become a necessity in life, mobile devices have created requirements for an enormous volume of multimedia transmissions. However, multimedia streaming is limited by the finite network bandwidth, the computation power of client processors, and the size of memory. There is a need for scalable video quality and flexible streaming mechanism that are able to adapt the frame processing rate in real time.

The Thesis proposes a Software-Defined Monitoring (SDM) design for fast network flow analysis. We separate the flow searching and packet dropping into three modules, namely parser, rule table, and executer. Compare to the pure software streaming method, we further adopt hardware parallelism for multimedia streaming by implementing SDM on a Network-on-Chip (NoC).

For video quality measurement, when the GOP size is 4, dropping only P frames in each GOP results in better video quality (33.45 dB) than dropping B frames (31.77 dB). When the length of a GOP is large, significant improvement video quality (PSNR) can be achieved 6 dB. For a smaller size of GOP, higher video quality can be achieved by packet dropping.

For comparison of different streaming methods, by dropping video frames, SDM can achieve a lower CPU utilization than the software streaming. The CPU utilization can be reduced to a maximum of 13%. Experiment results show that SDM provides better results compared with pure software streaming.

並列關鍵字

Network-on-Chip ； Software-Defined Monitoring ； Network Flow

參考文獻

[21] M. Forconesi, G. Sutter, S. Lopez-Buedo, and J. Aracil. Accurate and flexible

Conference on Image Processing (ICIP), pages 689–692, October 2008.

[2] L. Kekely, V. Pus, and J. Korenek. Software defined monitoring of application

protocols. In Proceedings of the 33rd Annual IEEE International Conference

on Computer Communications (INFOCOM’14) on Computer Communications,