mersenneforum.org > Data Proposed DC assignment and recycle rules
 Register FAQ Search Today's Posts Mark Forums Read

2014-01-21, 01:41   #12
Aramis Wyler

"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

35610 Posts

Quote:
 Originally Posted by chalsall If it's not hard, then may we please ask that this is implemented?
Seems unnecessary, have we lost sight of the big picture here? We don't need to turn our topiary into stumps.

Unrelated, I'm concerned about the reassignment of candidates who haven't been reported on in 60 days. I think that rule should probably be entirely struck (the other rules would eventually reassign those candidates) Because if a computer is offline for a couple months but completes the queued work, why would we reassign that? The rules about Assignments not having been completed for x amount of time and the rules about the expected completion date (calculated from the first update) taking > x amount of time should be sufficient.

2014-01-21, 02:25   #13
Prime95
P90 years forever!

Aug 2002
Yeehaw, FL

2·5·691 Posts

Quote:
 Originally Posted by Aramis Wyler Seems unnecessary, have we lost sight of the big picture here? We don't need to turn our topiary into stumps.
Are you referring to the chalsall's "has completed work recently" addition?

This would only apply to getting preferred assignments. The idea is bolster the already existing "speed is sufficient" criteria.

Quote:
 Unrelated, I'm concerned about the reassignment of candidates who haven't been reported on in 60 days. I think that rule should probably be entirely struck.
This rule is already in place - and has been since forever. I updated the original post with asterisks to indicate the rules that are already in place.

This rule does not apply to manual assignments. I'll update the first post.

 2014-01-21, 03:50 #14 TheMawn     May 2013 East. Always East. 6BF16 Posts I like the rules. I'll relax my aggressive expectations. Other users have made good cases for themselves. Regarding the 3000, I could see 6000/1500 being a plausible setup too. There's no particular reason to keep a 2:1 ratio that I can think of. The recycling rules are in place to clean up stuck assignments. The assignment rules are in place to stop users who are likely to get assignments stuck. If we're successful in keeping them out, the recycling rules are much less likely to ever get invoked. An assignment can be for exponent #5999 where the user is quite likely to finish on time. If they don't, the computer is likely down and because the amount of work itself is fairly low (especially since we're talking about fast-ish computers) it won't be a huge deal if a half-done assignment is cancelled. The assignment could also be for exponent #6001 which could apply to practically anyone. The bigger the gap between 6000 and 1500, the more time people get to finish. If the gap is too small, people may unknowingly grab an assignment which they don't actually have time to finish. 3000/1500 leaves however long it takes to do 1500 preferential assignments (which I assume will be handed out to any qualifying computer as soon as they are available). At 25 per day, that's two months. The user was promised 2 years as long as they update, but two months later, they're suddenly given a 30-day deadline. At 16 per day, that's 94 days. The user was promised two years but they suddenly step into top-1500 territory and their assignment is IMMEDIATELY recycled without warning. 6000/1500 triples the amount of time the user gets to do their medium-priority assignment but pretty much guarantees that as soon as their assignment reaches top-1500, they will immediately lose it. That's pretty much the only drawback to a larger gap. I don't know what the preferential DC rate is going to be, since we're giving it to more reliable workers, but the gap should be large enough to actually come close to giving people the two years they were promised. Unless we tighten that number a bit, too. I'm thinking even 9000/1500 is a better setup. EDIT: Or even 12000/1500. Last fiddled with by TheMawn on 2014-01-21 at 03:56
 2014-01-21, 04:00 #15 TheMawn     May 2013 East. Always East. 11×157 Posts Would it be too much work to add a third assignment category? My suggestions in red. Code: Assignment rules (top 3000 1500 to 12000 exponents): *1) No v4 users *2) No manual reservations *3) Reliability is high. *4) Speed is satisfactory. 5) "Days of work" setting is short. Assignment rules (top 1500 exponents): *1) No v4 users *2) No manual reservations *3) Reliability is high. *4) Speed is satisfactory meets first-time LL requirements. 5) "Days of work" setting is short. Recycle rules (top 1500 exponents): *1) If not a manual assignment and expected completion date is not updated for 60 days. 2) If assignment made before 2014-02-01 and: 2a) assignment is 12 months old and < 50% complete it is recycled. 2b) assignment is 15 months old it is recycled. 3) If assignment made after 2014-02-01 and: 3a) assignment is 6 months old and not started it is recycled. 3b) assignment is 9 months old it is recycled. 4) If DC is completed by another user, the assignment is cancelled. Assignment rules (not top 3000 12,000): *1) Speed is satisfactory. Recycle rules (not top 1500): *1) If not a manual assignment and expected completion date is not updated for 60 days it is recycled. 2) If assignment is one year old and not started it is recycled. 3) If assignment is two years old it is recycled. 4) If DC is completed by another user, the assignment is cancelled. EDIT: I want the gap to be big to prevent users from unknowingly stepping into an assignment they can't complete in time. That leaves 10,500 exponents to reliable users and the last 1,500 to really reliable users. The larger gap also reduces the risk of the exponent being recycled and given back to another troll user. Last fiddled with by TheMawn on 2014-01-21 at 04:04 Reason: Explanations
2014-01-21, 04:18   #16
Aramis Wyler

"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

22·89 Posts

Quote:
 Originally Posted by Prime95 Are you referring to the chalsall's "has completed work recently" addition?
No, I was refering to the 'computer has returned at least 2 * num_cores results in the last 3 months' rule mod. My cpu is old but has 4 cores, but sometimes I only use 3 for prime95 (I'm having some issues with windows 7 being sluggish when all 4 cores are running) I'm not sure I actually tend to get done 8 numbers per 3 months. almost all of the work I turn in is from my (2) video cards.

Thank you for the clarification about manual assignments.

 2014-01-21, 04:31 #17 Uncwilly 6809 > 6502     """"""""""""""""""" Aug 2003 101×103 Posts 22·2,053 Posts I don't think that DC machines need to meet the same performance standards as first time LL. As machines age, the same machine that was ok for LL for years should be stepped down to DC, then TF or LMH-TF. I am at 91% lifetime DC by using oldish machines and I haven't run a single DC in years (I think at least 6 years). This has been fairly standard for years.
2014-01-21, 05:52   #18
Prime95
P90 years forever!

Aug 2002
Yeehaw, FL

691010 Posts

Quote:
 Originally Posted by TheMawn I want the gap to be big to prevent users from unknowingly stepping into an assignment they can't complete in time.
This is IMO the major flaw with my proposal.

Your typical casual user installs prime95. After a few months their computer starts getting preferred assignments. Now for some reason the user changes his computer habits (such as running the PC part time because it is summer, the computer moved to a bedroom, the wife complained about electric bills, company policy change, etc). His computer faithfully works on the exponent and after 7 months turns in a result only to be told the result was not needed because Primenet expired the exponent after 6 months. Result: a pissed off user.

I'm tempted to say preferred DCs and LLs should be a separate work type. By selecting that work type, the user is volunteering to work on exponents that will be expired more quickly than standard exponents.

The downside is I think this would require a client change.

2014-01-21, 08:01   #19
tha

Dec 2002

24×72 Posts

Quote:
 Originally Posted by Prime95 The only real advantage to reclaiming exponents before they enter the critical top-2000 is that it lets users get slightly smaller assignments at the leading edge. Do you think the reclaim rules in the non-top-2000 are too aggressive? Recommended changes?
I think the main advantage to expand the top is that assignments enter it in an earlier period of time and therefore have the trailing end follow the preferred treshold in a smoother way.

I don't think the non-top-2000 is too agressive. The design philosophy behind it should be that we let everyone play as they want, unless it is really hopeless or if people claim too many assignments at the same time. A cap on the amount of assignments for the same computer might do more to smoothen the process

2014-01-21, 14:30   #20
petrw1
1976 Toyota Corona years forever!

"Wayne"
Nov 2006

2·17·127 Posts

Quote:
 Originally Posted by Prime95 4) If DC is completed by another user, the assignment is cancelled.
My concern is if I am making good progress on a DC and it gets poached then I lose all the work. Would it not be of value as a triple check if it is allowed to complete. Still needs a reasonableness check for % complete and days to done.

2014-01-21, 14:35   #21
petrw1
1976 Toyota Corona years forever!

"Wayne"
Nov 2006

10DE16 Posts

Quote:
 Originally Posted by Prime95 I'm tempted to say preferred DCs and LLs should be a separate work type. By selecting that work type, the user is volunteering to work on exponents that will be expired more quickly than standard exponents. The downside is I think this would require a client change.
Is this not more or less what GPU72 provides if you choose to join : prefered work for trusted contributers, policed agressively. However I agree this should be in P95 for those who can't or won't join G72.

2014-01-21, 19:32   #22
TheMawn

May 2013
East. Always East.

110101111112 Posts

Quote:
 Originally Posted by Uncwilly I don't think that DC machines need to meet the same performance standards as first time LL. As machines age, the same machine that was ok for LL for years should be stepped down to DC, then TF or LMH-TF. I am at 91% lifetime DC by using oldish machines and I haven't run a single DC in years (I think at least 6 years). This has been fairly standard for years.
Check my counter-proposal again. I was referring to top-1500 DC meeting first-LL standards. Only the top-1500. There's 10,500 other assignments that aren't also available to people running Prime95 on their microwave.

Quote:
 Originally Posted by Prime95 Your typical casual user installs prime95. After a few months their computer starts getting preferred assignments. Now for some reason the user changes his computer habits (such as running the PC part time because it is summer, the computer moved to a bedroom, the wife complained about electric bills, company policy change, etc). His computer faithfully works on the exponent and after 7 months turns in a result only to be told the result was not needed because Primenet expired the exponent after 6 months. Result: a pissed off user. I'm tempted to say preferred DCs and LLs should be a separate work type. By selecting that work type, the user is volunteering to work on exponents that will be expired more quickly than standard exponents.
I also think some sort of disclaimer would do the trick. Perhaps a checkbox that says "request preferred work if it is available and computer meets requirements?" Leave a note which says preferred assignments have stricter deadlines.

There will pretty much always be a problem on the edge of the transition from one level of preference to another. If someone grabs exponent #1501, an hour later it becomes exponent #1499 and suddenly runs into much stricter requirements as far as recycling is concerned. The solution is preventing people from getting #1501 who can't meet recycling day for #1499, but this can really only be done by adding an intermediate category, but this becomes complicated.

Perhaps there could be assignment rules top-3000, 3000-12000, not-top-12000 and then recycle rules top-1500 and not-top-1500.

Practically anyone can get #12001 but it has to become #1499 before the recycling rules are affected. A semi-decent user can get #3001 but it has to become #1499 before recycling rules are affected.

Is a gap of 1500 enough? We could increase it to 4500, but is it too much to reserve 6000 exponents for higher-yet standard machines?

Quote:
 Originally Posted by petrw1 My concern is if I am making good progress on a DC and it gets poached then I lose all the work. Would it not be of value as a triple check if it is allowed to complete. Still needs a reasonableness check for % complete and days to done.
This is why in the other thread I asked about poaching rules. I think the poacher should be denied credit and the poachee should be granted it immediately. Credit meaning GHz-Days credit and discovery credit.

 Similar Threads Thread Thread Starter Forum Replies Last Post S485122 PrimeNet 7 2018-06-08 14:49 Prime95 PrimeNet 74 2017-01-18 18:36 Fred PrimeNet 3 2016-05-19 13:40 Prime95 Data 156 2015-09-19 12:39 Prime95 Data 9 2014-02-27 23:52

All times are UTC. The time now is 19:02.

Thu Jul 9 19:02:17 UTC 2020 up 106 days, 16:35, 1 user, load averages: 1.96, 1.79, 1.84