High-performance Processing in Networked and Grid Environments

Abstract

In this dissertation, we present several techniques to achieve the high-performance processing in networked and grid environments. Many applications need a high-performance processing system to execute efficiently. High-performance processing mainly stems from parallelism. The parallel nature of grid computing is a very attractive solution to exploit the mentioned parallelism by executing either different parts of an application or several applications in parallel. In a grid system, the most important resources are computing and communication resources. The computing resources are the processors in the nodes on the grid. Communication within the grid is important for distributing tasks and their required data to the nodes within the grid. We propose an innovative high-performance platform to utilize reconfigurable processors in grid environments. Furthermore, we focus on the communication infrastructures and network processing (processing required for packets) platforms to utilize them through the grid environments. The collaboration of reconfigurable processors in a grid environment is presented and several compute-intensive multimedia kernels are simulated. Subsequently, we introduce three approaches to accelerate network processing tasks using Bloom filters in the networked and grid environments. The first and second techniques present a cache architecture for a counting Bloom filter (CCBF) and a memory optimization approach for Bloom filters using an additional hashing function (BFAH). The third technique proposes a power efficient pipelined Bloom filter. We present the results of our proposed approaches in collaboration of reconfigurable processors in grid computing (CRGC) and Bloom filters in network processing applications, e.g., packet classification. The experimental results show that the CRGC approach improves performance up to 7.2x and 2.5x compared to a GPP and the collaboration of GPPs, respectively. The results of the CCBF and BFAH for packet classification show that the proposed techniques decrease the number of memory accesses when compared to a standard Bloom filter.