execution time is long but the biggest issue is dataset size. Unoptimized code on main produces a dataset size of ~28 GB with SPEC benchmark 657 test dataset. Initial optimization was done by removing pid from each entry. This reduced set size down to ~20 GB. The next optimization is to pipe all output into bzip compression. This drastically reduces set size to ~4-5 GB but at the cost of increased execution time. The next optimization that I will try will be to implement a period parameter similar to what is implemented in the LaCASA mtrace program. http://lacasa.uah.edu/index.php/software-data/mtrace-tools-and-traces
Another idea for decreasing trace size is to implement a granularity parameter that reduces the need to have a timestamp (64-bits) associated with each instruction in the trace.
execution time is long but the biggest issue is dataset size. Unoptimized code on main produces a dataset size of ~28 GB with SPEC benchmark 657 test dataset. Initial optimization was done by removing pid from each entry. This reduced set size down to ~20 GB. The next optimization is to pipe all output into bzip compression. This drastically reduces set size to ~4-5 GB but at the cost of increased execution time. The next optimization that I will try will be to implement a period parameter similar to what is implemented in the LaCASA mtrace program. http://lacasa.uah.edu/index.php/software-data/mtrace-tools-and-traces