(archive)MISFIT

[swl551]

Originally Posted by Chuck

I remember that there is a problem with Primenet's parsing of results if the first result submitted is a "factor found"; Primenet considers it to be a P-1 result if there are no preceeding "no factor" lines in the submission.

Does MISFIT do any reordering of results submitted to avoid this?

Yes it does.

[swl551]

Version 2.4.4
1. Added purging of all MISFIT generated logs. Checking is performed roughly every six hours and is set and forget (cannot be triggered manually). Configured in the MISC tab. **GPU72FETCH_logs which is not part of MISFIT and not purged

2. Added global exponent duplicate detection report. Detection is for everything in inventory, workToDo(s) and MISFITworkToDo files.

3. Added configuration to assess bit level as part of any duplicate detection procedure. It can turned on in the config|MISC tab.
If bit level evaluation is turned on the following example assignments would NOT be considered as having duplicate exponents
55555544,71,72
55555544,72,73
55555544,72,75
55555544,75,45

If bit level evaluation is enabled it will impact the results of the following procedures:
a. global exponent duplicate report
b. detecting duplicates during GridStatsUpdate
c. detecting and tossing duplicates during WorkBalance

Get from http://mersenneforum.org/misfit/

[kladner]

Sounds great. You can probably disregard most of the PM which I cross-posted with yours above.

[kracker]

Hell yeah, the purge actually works now!.... *poke poke*

Thanks.

EDIT: the above to swl551.

[chalsall]

Putting this out there for discussion...

Since Scott has done such a wonderful job of "taking the human out of the loop", would anyone object if I remained in the loop?

Specifically, I'm wondering if anyone would object to the idea that when GPU72 sees a request for assignments from a MISFIT-TF-FETCHER client with the "What Makes Sense" option set, that the system ignores all the other settings (i.e. Range and TF level), and issues assignments (of the worktype requested; as in if LLTF is requested it will be issued, same with DCTF) which makes the most sense for GIMPS as a whole?

If any other option was set (i.e. Lowest TF level, etc.) the system would honor the request explicitly as best it could.

The reason I ask is there is a chance we won't have to release at 72 -- although I very much appreciate everyone's diligence in building a buffer to there in case we do need to.

But if the "surge" subsides and we have the time to go to 73, rather than asking everyone to change (once again) their MISFIT settings, I could simply make one change in one Perl script and everyone using MISFIT would be doing what really "Makes Sense"^(TM).

Thoughts?

[swl551]

Originally Posted by chalsall

Putting this out there for discussion...

Since Scott has done such a wonderful job of "taking the human out of the loop", would anyone object if I remained in the loop?

Specifically, I'm wondering if anyone would object to the idea that when GPU72 sees a request for assignments from a MISFIT-TF-FETCHER client with the "What Makes Sense" option set, that the system ignores all the other settings (i.e. Range and TF level), and issues assignments (of the worktype requested; as in if LLTF is requested it will be issued, same with DCTF) which makes the most sense for GIMPS as a whole?

If any other option was set (i.e. Lowest TF level, etc.) the system would honor the request explicitly as best it could.

The reason I ask is there is a chance we won't have to release at 72 -- although I very much appreciate everyone's diligence in building a buffer to there in case we do need to.

But if the "surge" subsides and we have the time to go to 73, rather than asking everyone to change (once again) their MISFIT settings, I could simply make one change in one Perl script and everyone using MISFIT would be doing what really "Makes Sense"^(TM).

Thoughts?

I see the idea as very high value since we all want to do what is right.
Another option would be the LLTF and DCTF pages accept an option value equating to "Mission Specialty" this would allow the "Fallback to what makes sense" still viable if the Mission Specialty ran dry or was cancelled/completed.

[chalsall]

Originally Posted by swl551

I see the idea as very high value since we all want to do what is right.

That's not entirely true -- most want to do what is best for GIMPS, but some want to do very specialized work. Particular ranges; find the most factors, etc. All work is valuable, so in the long run all work is good for GIMPS, but sometimes some want to do what doesn't make the most sense at this particular moment in time.

Since this is a volunteer effort, everyone should be allowed to do whatever they want with their own electricity, hardware and time (so long as there's no negative impact on the project or any other participant).

Originally Posted by swl551

Another option would be the LLTF and DCTF pages accept an option value equating to "Mission Specialty" this would allow the "Fallback to what makes sense" still viable if the Mission Specialty ran dry or was cancelled/completed.

That's what I envisioned all the other options are.

The issue is I have sort of painted myself into a corner. The web-interface was never intended to be spidered. As in, before MISFIT there was a human in the loop who could see if the options they were specifying (Option, Ranges, Pledge level) resulted in no available assignments.

If I had been smarter, the "What Makes Sense" option would have always ignored all the other settings. But, as has been demonstrated, I'm not that smart...

This is why I initially discouraging you from using the "intended for a human in the loop" interface. I had hoped to be able to expose an API appropriate for automation. But then life (and a death) happened, and I realized there was a serious vacuum which needed to be filled and you were keen to fill it.

So the question at hand is how do we best deal with what we have available right now. Specifically, I'm asking if anyone would mind if the "What Makes Sense" behavior changed when a human isn't in the loop.

Edit: Unlike Micro$oft, I don't like changing the behavior of things without notice.

[Aramis Wyler]

I'd rather see this as a 5th dropdown option in gpu72, such that it affects all users and all methods, and not something specific to misfit.

I think that was clear, but to be extra repetitive, I'd like to see a 5th option on the dropdown on the gpu72 web site that says 'Do what Chris wants' or the like (ha!) and nothing on the form matters (but you get the preview).

There would be no change to the workfetcher or misfit (except that '5' or '0' would be a legal option).

[chalsall]

Originally Posted by Aramis Wyler

I'd rather see this as a 5th dropdown option in gpu72, such that it affects all users and all methods, and not something specific to misfit.

Sorry... I should have been clearer...

To any human accessing the Assignment forms there should be no changes needed, since I can already update the default settings which the human sees (and the preview shows the results of) as the real-time situation requires.

My proposal is only for MISFIT users, who only yesterday had to change their "pledge" setting from 73 to 72 even though most already had "What Makes Sense" as the option.

Again, this is actually a mistake I made long ago -- WMS should have always ignored all the other settings.

[swl551]

MISFIT and the stand-alone workFetcher both pass in the same UserAgent string. I can make them unique if required.

[chalsall]

Originally Posted by swl551

MISFIT and the stand-alone workFetcher both pass in the same UserAgent string. I can make them unique if required.

Thanks, but no need.

I've seen a couple of users using human-driven browsers which didn't provide a UA string. With the MISFIT... UA string I know a human isn't in the loop, and the system can act accordingly.

I won't make the proposed change until we hear back from some more humans.