Abuse (?) of the manual assignment system

S485122 · 2023-02-21, 10:35

This has been discussed previously. But it seems as if a person found pleasure in abusing the Manual Assignment system.

Code:

Range    Type     # %/total
    2     ECM  2724   59,2%
    3     ECM  3352   66,1%
    4     ECM   609   53,8%
   68     PRP   52
   82     PRP   48
  100     PRP   96
  115     PRP   36
  116   PM1-L   14
  117   PM1-L 16149   99,9%
  118   PM1-L 14011   98,5%
  119   PM1-L  9919   91,5%
  255      TF     2
  332     PRP    33
  388     PRP     1

On 2022-12-11 a user took about 40 000 PM-1 assignments in the 117M, 118M and 119M ranges, representing more than 97 % of that type and almost 70% of all non factored Mersennes in those ranges.

That same user took 200 PRPs assignments a few days earlier on 2022-11-30.

And that user has still not returned a single result (unless they are returned under another ID ?)

It wouldn't be a problem, since there is sufficient work remaining assignable ... but if gives the illusion that enough power is committed on P-1 and it blocks work for others in those leading edge ranges.

Andrew Usher · 2023-02-21, 11:51

Wow. The P-1 assignments, both in number and in importance, make this considerably worse than I previously thought (https://www.mersenneforum.org/showthread.php?t=28379). I can't say for sure what his intent was, but it's clear that this is a huge hole in the assignment system.

All of the assignments of "PRAMEY PANDEY ", in all types, ought to be expunged. And unfortunately we have to be prepared for this to possibly be tried again.

James Heinrich · 2023-02-22, 15:37

User is banned from manual assignments and all assignments expired.

kruoli · 2023-02-22, 16:06

It could be a possibility to limit the number of assignment a user without recent work can have. As soon as he has returned recent work, he is allows to assign more, but no more than x time of the work returned recently.

Uncwilly · 2023-02-22, 16:10

Maybe only as much work as one has previously returned in the last 365 days. (Maybe with a default factor for starting off that assumes 1 mid level CPU and 1 mid level GPU.)

James Heinrich · 2023-02-22, 16:35

There's a whole pile of issues around limiting things like that. It's extremely hard to judge what a "reasonable" amount of work is -- a new user might join with a single quad-core CPU, or the new user might be Ryan Propper or similar. Don't want to artificially limit new high-power users and discourage them.

I'll try poking around later today and see if can come up with a report of some sort to identify potential too-many-assignment users and if something should be done about it.

kriesel · 2023-02-22, 16:55

Or the user might suddenly be requesting a lot more assignments, because of expanding hardware.
That would be a natural outcome of adding a Radeon VII or two, or an RTX4090, or a Xeon Phi.
The Phi can run prime95 or mprime, but GPUs tend not to be using the PrimeNet API.

James Heinrich · 2023-02-22, 20:18

Quote:

Originally Posted by James Heinrich

see if can come up with a report of some sort to identify potential too-many-assignment users

Here now:
https://www.mersenne.org/assignments/topusers.php

SRBase is well-known and has official blessing to have vast numbers of assignments and results.
I'm not sure if GPU Factoring = GPU72

, if so that also makes sense.

Mark Rose · 2023-02-22, 20:19

Or when I brought my 8 3070s online.

It's not illegitimate to fetch thousands of TF assignments.

Or when I got a half dozen quad cores running again, doing only P-1.

I've currently got 887 P-1 assignments (not in front of the wavefront). I complete a few dozen a day.

kriesel · 2023-02-22, 21:23

For a while, I was running 40 P-1/day/GPU times up to 4 GPUs. And a typical queue of 1200 assignments total, mostly wavefront one-or-two-bit-elevation TF, was mostly only a couple days latency. (Currently a lot of that throughput is applied to OBD and MM127 so does not appear in mersenne.org stats.)

Andrew Usher · 2023-02-23, 00:55

Yes, those are entirely reasonable amounts to have. Also, we can and should be more lenient with TF on high exponents - any progress at all justifies assignments up there.

'GPU Factoring' is in fact GPU72, I've been told by someone that would know; it makes sense that they'd be holding a lot of exponents.

We ordinary users can't search for assignments by user name (perhaps that could be added?), but of the high-assignment list you might want to check out 'kirkland' whom I don't remember seeing any results from; also 'Bill' was the other name I mentioned in my original hoarding thread as in the same situation as the other user and possibly the same person. No one else looks at all suspicious.

Thanks for addressing the issue now, as to how to prevent it from arising again - the suggestion was made above to limit assignments to the number returned in the past year, which is generous; I'd endorse that, with a minimum for new users of (say) 100, and presumably administrators allowed to exempt a few trusted users. Breaking it down by type of assignment would be just as good but more complicated; I see no need to distinguish manual and non-manual assignments.

2023-02-21, 10:35	#1
S485122 "Jacob" Sep 2006 Brussels, Belgium 3×647 Posts	Abuse (?) of the manual assignment system This has been discussed previously. But it seems as if a person found pleasure in abusing the Manual Assignment system. Code: Range Type # %/total 2 ECM 2724 59,2% 3 ECM 3352 66,1% 4 ECM 609 53,8% 68 PRP 52 82 PRP 48 100 PRP 96 115 PRP 36 116 PM1-L 14 117 PM1-L 16149 99,9% 118 PM1-L 14011 98,5% 119 PM1-L 9919 91,5% 255 TF 2 332 PRP 33 388 PRP 1 On 2022-12-11 a user took about 40 000 PM-1 assignments in the 117M, 118M and 119M ranges, representing more than 97 % of that type and almost 70% of all non factored Mersennes in those ranges. That same user took 200 PRPs assignments a few days earlier on 2022-11-30. And that user has still not returned a single result (unless they are returned under another ID ?) It wouldn't be a problem, since there is sufficient work remaining assignable ... but if gives the illusion that enough power is committed on P-1 and it blocks work for others in those leading edge ranges. Last fiddled with by S485122 on 2023-02-21 at 10:58 Reason: more conditional...

2023-02-22, 21:23	#10
kriesel "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 17056₈ Posts	For a while, I was running 40 P-1/day/GPU times up to 4 GPUs. And a typical queue of 1200 assignments total, mostly wavefront one-or-two-bit-elevation TF, was mostly only a couple days latency. (Currently a lot of that throughput is applied to OBD and MM127 so does not appear in mersenne.org stats.) Last fiddled with by kriesel on 2023-02-22 at 21:27

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Manual Assignment: Not Needed?	Magellan3s	PrimeNet	8	2022-04-30 17:22
Manual assignment bug?	greg	PrimeNet	3	2020-05-07 16:58
Manual assignment	Unregistered	Information & Answers	12	2010-10-07 16:15
Problem with LL manual assignment	JuanTutors	Software	2	2008-12-02 01:57
Manual reservation system obsolete? (Answer: no)	thommy	Prime Sierpinski Project	5	2006-02-21 16:36

2023-02-21, 11:51	#2
Andrew Usher Dec 2022 2⁶×7 Posts	Wow. The P-1 assignments, both in number and in importance, make this considerably worse than I previously thought (https://www.mersenneforum.org/showthread.php?t=28379). I can't say for sure what his intent was, but it's clear that this is a huge hole in the assignment system. All of the assignments of "PRAMEY PANDEY ", in all types, ought to be expunged. And unfortunately we have to be prepared for this to possibly be tried again.

2023-02-22, 15:37	#3
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 2²×1,061 Posts	User is banned from manual assignments and all assignments expired.

2023-02-22, 16:06	#4
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 617₁₆ Posts	It could be a possibility to limit the number of assignment a user without recent work can have. As soon as he has returned recent work, he is allows to assign more, but no more than x time of the work returned recently.

2023-02-22, 16:10	#5
Uncwilly 6809 > 6502 """"""""""""""""""" Aug 2003 101×103 Posts 73·151 Posts	Maybe only as much work as one has previously returned in the last 365 days. (Maybe with a default factor for starting off that assumes 1 mid level CPU and 1 mid level GPU.)

2023-02-22, 16:35	#6
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 2²×1,061 Posts	There's a whole pile of issues around limiting things like that. It's extremely hard to judge what a "reasonable" amount of work is -- a new user might join with a single quad-core CPU, or the new user might be Ryan Propper or similar. Don't want to artificially limit new high-power users and discourage them. I'll try poking around later today and see if can come up with a report of some sort to identify potential too-many-assignment users and if something should be done about it.

2023-02-22, 16:55	#7
kriesel "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 1111000101110₂ Posts	Or the user might suddenly be requesting a lot more assignments, because of expanding hardware. That would be a natural outcome of adding a Radeon VII or two, or an RTX4090, or a Xeon Phi. The Phi can run prime95 or mprime, but GPUs tend not to be using the PrimeNet API.

2023-02-22, 20:19	#9
Mark Rose "/X\(‘-‘)/X\" Jan 2013 https://pedan.tech/ 2×3³×59 Posts	Or when I brought my 8 3070s online. It's not illegitimate to fetch thousands of TF assignments. Or when I got a half dozen quad cores running again, doing only P-1. I've currently got 887 P-1 assignments (not in front of the wavefront). I complete a few dozen a day.

2023-02-23, 00:55	#11
Andrew Usher Dec 2022 1C0₁₆ Posts	Yes, those are entirely reasonable amounts to have. Also, we can and should be more lenient with TF on high exponents - any progress at all justifies assignments up there. 'GPU Factoring' is in fact GPU72, I've been told by someone that would know; it makes sense that they'd be holding a lot of exponents. We ordinary users can't search for assignments by user name (perhaps that could be added?), but of the high-assignment list you might want to check out 'kirkland' whom I don't remember seeing any results from; also 'Bill' was the other name I mentioned in my original hoarding thread as in the same situation as the other user and possibly the same person. No one else looks at all suspicious. Thanks for addressing the issue now, as to how to prevent it from arising again - the suggestion was made above to limit assignments to the number returned in the past year, which is generous; I'd endorse that, with a minimum for new users of (say) 100, and presumably administrators allowed to exempt a few trusted users. Breaking it down by type of assignment would be just as good but more complicated; I see no need to distinguish manual and non-manual assignments.