透過您的圖書館登入
IP:3.135.183.1
  • 學位論文

加速電腦網路通訊協定訊息傳送效率之緩衝器管理架構

An Efficiency Buffer Management Architecture for Inter-Layer Message Passing in Communication Protocols

指導教授 : 鍾添曜
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


傳統上改善通訊協定效率的主要方法, 大都著手在減少層與層間的資料拷 貝, 然而它們並沒有考慮到網路界面接收資料時, 網路驅動程式與 OS 間 互動所發生的額外負擔, 這種額外負擔的主要來源是網路界面拷貝資料 時, 網路驅動程式要求 OS 配置記憶體的時間消耗, 為解決此問題, 本論 文提出『加速電腦網路通訊協定訊息傳送效率之緩衝器管理架構( Efficiency Buffer Management Architecture, EBMA)』。 在 EBMA 中 緩衝器監督器(Buffer Monitor, BM) 會預先配置一組記憶體給緩衝器池( Buffer Pool, BP) , 接收網路界面傳來的資料。 當網路界面接收到資料 封包而產生中斷時, BM 可以直接由 BP 取出記憶區塊使用, 不須要求 OS 配置記憶體, 如此可以減少要求 OS 配置記憶體的時間消耗。為了改善記 憶體的使用率, 我們使用監督事件計時器(Monitor Event Trigger, MET) 觸發 BM 來監督 BP 中預置緩衝區塊的數量。 當應用程式釋放記憶體時, EBMA 可將其歸還給 BP 循環使用, 而不直接歸還 OS, 因此可以減少從系 統配置記憶體的頻率, 我們使用模擬的方式充分了解 13 種 EBMA 模型, 並在一個專屬系統 (embedded System) 上實作 EBMA 來驗證其效能, 模 擬與實驗結果顯示, 使用 EBMA 製作的網路傳輸設備操作效率, 比不使用 EBMA的系統好。

並列摘要


Traditionally performance improvements in protocol suits implementation tried to reduce data copy between each layer. However, they do not consider the overhead caused by the operating system interaction while data is received from the network interface. This overhead is mainly due to the memory allocation before data is copied from the network interface. To solve this problem, this thesis proposes an Efficiency Buffer Management Architecture(EBMA). In EBMA a Buffer Monitor Module( BM) pre-allocates a bank of Buffer Pool(BP) for incoming data from the network interface. When an interrupt is generated for an incoming data packet, BM will pass a buffer to the receiving process without interacting with the operating system. This significantly reduces the overhead caused by memory allocation from the system. To improve memory utilization, we use a buffer Monitor Event Timer (MET) to triggers BM to monitor the size of the pre-allocated buffers. When buffers are released from the applications, instead of returning the released buffers to the system directly, the released buffers are returned to BP. This can reduce memory allocation from the OS . We use simulation to fully explore thirteen EBMA models. Furthermore, we also implement EBMA in an embedded system to verify its performance. Simulation and experiments results show that the proposed EBMA out performs existing systems significantly.

延伸閱讀