利用圖形處理器加速3-DES運算

Abstract

加密與解密計算至今已經被開發出各式各樣的演算法，針對不同使用上的需求提供不同的計算方式，已經幾乎成為網路傳輸中不可或缺的重要角色。然而，加密與解密的過程是需要龐大的CPU計算時間和資源，對一個支援加密傳輸的web server而言，往往web server本身會花60%至70%的資源在執行加解密的計算，對執行效率來說是很大的瓶頸。本研究探討這限制之最根本的原因，就是加解密過程耗費過多CPU資源，因此提出使用GPU（圖形處理器）來加速這些加解密的計算，以減少CPU的負擔，讓CPU可以更專注於其他web service的服務。 本論文並非是目前為止第一個利用了GPU的計算能力，來加速加解密運算之研究。然而加解密的密碼文件(cipher)種類相當多，我們針對計算量相當龐大的3-DES演算法進行研究，將用來原本該在CPU上之邏輯運算，轉換成可以在GPU上平行處理的演算法。在我們的實作中，隨著檔案大小的增加，我們觀察到在GPU提供了超過5倍目前CPU所能提供之運算能力，並大幅減少CPU所耗損的資源，有效的提供web server更好的服務品質。

Various cryptography algorithms have been developed to date to provide different levels of data security for application domains, such as storage security, personal identification, and secure web browsing. Although these algorithms do a very good job of protecting your privacy, they consume massive amount of resource on the server-side while processing encrypting and decrypting requests from clients. Generally speaking, a web server supporting transport security (TLS/SSL) could spend up to 60%~70% of its computing resource in encrypting and decrypting data and leave 30%~40% for actual client request processing. Therefore in this research, we try to address the performance issue by using GPU (Graphics Processing Unit) to speed up the data encryption and decryption to reduce the computing resource spent on security and ultimately improve the web server throughput. In this paper, we chose the widely-used 3-DES and implemented it on GPU. In our implementation, we observed the GPU cipher performs 5 times faster than the OpenSSL implementation on CPU. As a result, we show a promising direction for offloading the data encryption and decryption onto GPU.