Closed saarthdeshpande closed 1 year ago
Only linux_mem fails or all variants fail? Is this failure very reproducible? Could you run the experiment with ulimit -c unlimited to dump the core and post the stack frame here?
All variants throw a segmentation fault, but generate log files nonetheless. However, the exact results are not reproduced (attaching a comparison)
Wrt the stack frame: what ip_addr:port
do I pass as argv[2]
when running gdb main
?
oh okay, it's a benigh/intentional segfault as the experiment script just forcibly kill the process once it gets the result; in other words, you're safe to ignore it once you get the result log. In terms of result discrepency, are you using exactly the original parameters when running the experiment? You mentioned something like small input in your initial post?
Oh okay, great.
Yes, I'm using the original parameters. I mentioned small input because the seg fault is thrown only when kNumMutatorThreads
> 1. But since you said this is intentional, shouldn't be an issue.
Could you share how I can better reproduce the experiment results?
I believe the results are the same if you use the same yaxis scale with my figure
Thanks! I converted the cycles to microseconds as mentioned in the paper (2.4 GHz clock speed, 10 cores). The results match except for the last datapoint
great, it's normal if you see slight differences across runs as there can be many sources of randomness.
Great, thanks so much for your help! Closing this issue.
Executing run.sh in each subdirectory of aifm/exp/fig6a/ throws a segmentation fault.
For linux_mem, execution is error-free only when kNumMutatorThreads = 1. For all other values of kNumMutatorThreads, a segmentation fault occurs.
There is no other error message printed. How can I fix this?
Execution is on cloudlab using the profile provided in this repository.