What are the performance expectations for project 2?
First, note robustness and functionality are paramount. You must pass all basic tests (including test6 that tests full memory reclamation) to get any performance points at all.
We'll be using MSL 8, 16, 32; QS 8, 16, 32 (both large); NQueens 14 with 32 threads; and Fibonacci 41 with 32 threads.
For each benchmark, we'll be giving points like this:
- 100% - good performance
- 50% - mediocre performance
- 0% - performance lacking
Based on the scoreboard and my own implementation(s), we'll be using the following cutoffs (in seconds):
| Test | Good | Mediocre | Lacking | Serial (estimated) |
|---|---|---|---|---|
| MSL/8 | <=9 | 9-30 | >30 | 37 |
| MSL/16 | <=7 | 7-30 | >30 | 37 |
| MSL/32 | <=6 | 6-30 | >30 | 37 |
| QS/8 | <=8 | 8-30 | >30 | 36 |
| QS/16 | <=6 | 6-30 | >30 | 36 |
| QS/32 | <=5 | 5-30 | >30 | 36 |
| NQ/32 | <=7 | 7-30 | >30 | 124 |
| Fib/32 | <=5 | 5-60 | >60 | 15 |
Note that "Good" performance for all but Fib/32 is achievable with a single lock implementation.