Effects of Partitioning and Scheduling Sparse Matrix Factorization on Communication and Load Balance
INSTITUTE FOR COMPUTER APPLICATIONS IN SCIENCE AND ENGINEERING HAMPTON VA
Pagination or Media Count:
We present a block-based, automatic partitioning and scheduling methodology for sparse matrix factorization on distributed memory systems. Using experimental results, we analyze this technique for communication and load imbalance overhead. To study the performance effects, we compare these overheads with those obtained from a straightforward wrap-mapped column assignment scheme. All experimental results were obtained using test sparse matrices from the Harwell-Boeing data set. The results show that there is a communication and load balance trade-off. The block-based method results in lower communication cost whereas the wrap-mapped scheme gives better load balance.
- Administration and Management
- Theoretical Mathematics