|
| 1 | +# dmclock benchmarking |
| 2 | + |
| 3 | +**IMPORTANT**: now that K_WAY_HEAP is no longer allowed to have the |
| 4 | +value 1, the shell and Python scripts that generate the PDFs no longer |
| 5 | +work exactly correctly. Some effort to debug is necessary. |
| 6 | + |
| 7 | +This directory contains scripts to evaluate effects of different |
| 8 | +branching-factors (k=1 to k=11) in the IndirectIntrusiveHeap |
| 9 | +data-structure. IndirectIntrusiveHeap is now a k-way heap, so finding |
| 10 | +an ideal value for k (i.e., k=2 or k=3) for a particular work-load is |
| 11 | +important. Also, it is well-documented that the right choice of |
| 12 | +k-value improves the caching behaviour [Syed -- citation needed |
| 13 | +here]. As a result, the overall performance of an application using |
| 14 | +k-way heap increases significantly [Syed -- citation needed here]. |
| 15 | + |
| 16 | +A rule of thumb is the following: |
| 17 | + if number of elements are <= 6, use k=1 |
| 18 | + otherwise, use k=3. |
| 19 | + |
| 20 | +## Prerequisites |
| 21 | + |
| 22 | +requires python 2.7, gnuplot, and awk. |
| 23 | + |
| 24 | +## Running benchmark |
| 25 | + |
| 26 | +./run.sh [name_of_the_output] [k_way] [repeat] # [Syed -- last two command line args do not work] |
| 27 | + |
| 28 | +The "run.sh" script looks for config files in the "configs" directory, |
| 29 | +and the final output is generated as |
| 30 | +"name_of_the_output.pdf". Internally, "run.sh" calls other scripts |
| 31 | +such as data_gen.sh, data_parser.py, and plot_gen.sh. |
| 32 | + |
| 33 | +## Modifying parameters |
| 34 | + |
| 35 | +To modify k-value and/or the amount of times each simulation is |
| 36 | +repeated, modify the following two variables in "run.sh" file: |
| 37 | + |
| 38 | + k_way=[your_value] |
| 39 | + repeat=[your_value] |
| 40 | + |
| 41 | +For example, k_way=3 means, the benchmark will compare simulations |
| 42 | +using 1-way, 2-way, and 3-way heaps. |
0 commit comments