User talk:Thundergnat: Difference between revisions
→Anti-primes - concurrency failure?: Maybe need to use atomic-assign?
SqrtNegInf (talk | contribs) |
Thundergnat (talk | contribs) (→Anti-primes - concurrency failure?: Maybe need to use atomic-assign?) |
||
Line 385:
::Kind of went down the rabbit hole on this one... Stress-tested all of the RC tasks that use concurrency (mainly in an Ubuntu VM with the latest P6, but also Windows/2018.10). P6 performed pretty well (aside from sporadic crashes), always giving correct results '''except''' for 'Anti-primes'. So it doesn't appear to matter which OS or P6 version, in line with the race condition hypothesis. The best way I found to trigger is low free memory. Also, the story with 'Proper divisors' and .race: Always ended up with the right answer, but if you look at @candidates you see that it's different with each run. --[[User:SqrtNegInf|SqrtNegInf]] ([[User talk:SqrtNegInf|talk]]) 15:59, 11 January 2019 (UTC)
:::I ''still'' haven't seen a failure on my system running either of those. I may need to roll up a memory limited VM to see if it fails then. With 32GB of main memory I don't run into low memory situations very often.
:::I changed the hyper grep to use atomic-assign. That's the only place I can see where concurrency may be an issue. It continues to work flawlessly for me so I can't really judge if it helps or not. Maybe you could give the new code a try? If it is still a problem, I'll just remove the concurrency. It is fast enough without it. It's just nice to get the "free" speed-up. --[[User:Thundergnat|Thundergnat]] ([[User talk:Thundergnat|talk]]) 17:24, 12 January 2019 (UTC)
| |||