An FPGA architecture for the Pagerank eigenvector problem
Séamas McGettrick, Dermot Geraghty, Ciarán McElroy · 2008
Googlepsilas PageRank (PR) eigenvector problem is the worldpsilas largest matrix calculation. The algorithm is dominated by Sparse Matrix by Vector Multiplication (SMVM) where the matrix is very sparse, unsymmetrical and unstructured. The computation presents a serious challenge to general-purpose processors (GPP) and the result is a very lengthy computation time. In this paper, we present an architecture for solving the PR eigenvalue problem on the Virtex 5 FPGA. The architecture is optimised to take advantage of the unique features of the PR algorithm and FPGA technology. Performance benchmarks are presented for a selection of real Internet link matrices. Finally these results are compared with equivalent GPP implementations of the PR algorithm.