Performance Evaluation ofSparse Storage Fornats

Sparse matrices arepervasive inmanyComputational Science andEngineering (CS&E) applications. There isasignificant number ofstorage formats usedtorepresent sparse matrices. Thispaperpresents a performance evaluation ofstorage formats forthemainkernel ofiterative methods fornumerical linear algebra, namely matrix-vector multiplication. Theexperiments consider asetofalmost 200sparse matrices fromtheMatrix Market collection covering both systems oflinear equations andeigenvalue problems. For eachmatrix, theexperiments perfonn thematrix-vector multiplication withmostcommonly usedsparse storage formats andalso therecently proposed JavaSparse Array storage fonnat. Tothebest oftheauthors' knowledge, there isnoother performance evaluation ofstorage formats forsparse matrices which consider such avariety ofmatrices andstorage formats.