I'm trying to running the Freqmine on alpha full system using ruby_fs.py with 64 cores & cache options.
The first trial was successful, and it took 117 hours to complete the benchmark. At the second trial, I ran the benchmark with checkpoint (which
created at the first trial) restore option, but it has terminated with segmentation fault after running for 66 hours.
Here is terminal output for second trial.
KS] Entering ROI
transaction number is 250000
freqmine(1297): unaligned trap at 00000001200066c4: 000002002a95ee14 2d 16
freqmine(1296): unaligned trap at 00000001200066c4: 000002002955ae14 2d 16
freqmine(1299): unaligned trap at 00000001200066c4: 000002002d166e14 2d 16
freqmine(1298): unaligned trap at 00000001200066c4: 000002002bd62e14 2d 16
The terminal output from the first trial also has same ‘unaligned trap’ warning messages, but numbers were different (13XX). Do these warning
with some specific numbers cause segmentation fault?
I have search and read several posts regarding this problem and realized that using old-disk image file can cause segmentation fault. I'm using
pre-built disk image which comes with SPLASH2 and PARSEC benchmarks, but I’m not sure which version it is. Is there a way to check a version of disk image?
Is there anyone who completes this benchmark without segmentation fault?
Any help would be greatly appreciated!