Newer milestone thread - Page 38

garo · 2012-02-21, 22:09

I can testify that there have been at least 4 occasions since I joined GIMPS when George manually released exponents that had been reserved for too long. I haven't seen one in at least a couple of years though.

@Brian-E: Your supposition is incorrect. I for one had some exponents that I had been working on poached by aketilander while my main rig was down due to hardware issues. I sent him a PM but he said he was going to poach anyway.

I can only hope that with more time on the project he is less eager to poach.

garo · 2012-02-21, 22:11

Quote:

Originally Posted by chalsall

Here is an interesting example demonstrating that perhaps PrimeNet is not sane: 25195613.

Hmmmm....

No this happens when an exponent is reserved and report completion dates in Prime95 but is subsequently extended uses the manual form. The expected completion date is not updated.

cheesehead · 2012-02-21, 22:13

Quote:

Originally Posted by chalsall

cheesehead... Step back a bit...

What is being talked about here is "poaching" DCs. Those which have been reserved by those who do not appear to be serious,

... and you've decided, in your magnificence, that you have an accurate evaluation of the seriousness of others about whom you know only a few numbers?

I still haven't seen you offer any case which could not be handled by means other than poaching.

Quote:

and are holding up milestones.

(I'm considering writing a short paper entitled, "Milestones considered harmful" and submitting it to some distributed computing journal.

That's a small ACM joke.)

Hmmm ... was it you who recently posted something about "101 km/h"? Kilometerstones, perhaps?

Quote:

Yes, there is a VERY small possibility that a DC will find a MP which was missed by the first LL.

So, can you confirm that you at least have been persuaded never to poach a first-time LL assignment? And that from here on, we can discuss only DC poaching?

chalsall · 2012-02-21, 22:18

Quote:

Originally Posted by cheesehead

Do you really not get it -- about the anticipation and excitement of watching an LL test end?

I do get it.

Quote:

Originally Posted by cheesehead

So, communicate with GIMPS administrators. Wake up the old discussion threads about devising a more automated response to such situations.

I have reason to believe (or, at least, hope) that this discussion has done exactly that....

bcp19 · 2012-02-21, 22:19

How difficult would it be, on assignments over 1 year old, to do something like add another column or two to the database, and store % complete and date that is not overwritten for say 30 days, to track progress? Thus today, it would store 2/21/12 and 14.6%. If the assignment completes, it's obviously no longer tracked, but after say 30 days, it could check progress and if less than a certain % cancel the assignment. I don't know if P95 has the programming to toss an assignment if flagged by the server, but it seems like something to consider adding.

chalsall · 2012-02-21, 22:25

Quote:

Originally Posted by cheesehead

So, can you confirm that you at least have been persuaded never to poach a first-time LL assignment? And that from here on, we can discuss only DC poaching?

I can confirm that I have never "poached" a LL nor a DC assignment.

chalsall · 2012-02-21, 23:03

Here's an idea (which has been suggested before)...

Perhaps those who are "Anonymous" should be assigned work well ahead of the "waves".

cheesehead · 2012-02-21, 23:18

Quote:

Originally Posted by chalsall

Here's an idea (which has been suggested before)...

Perhaps those who are "Anonymous" should be assigned work well ahead of the "waves".

Now, _that_ could be a good idea!!

bcp19 · 2012-02-21, 23:31

Quote:

Originally Posted by chalsall

Here's an idea (which has been suggested before)...

Perhaps those who are "Anonymous" should be assigned work well ahead of the "waves".

Then again, why limit it to Anonymous? Isn't there currently a 'trust' level? If you had say 5 trust levels (anon = 5), Level 1 could get from lowest (x) to x+2M exp, level 2 from x+2M to x+4M, level 3 from x+4M to x+7M, level 4 from x+7M to x+11M and level 5 x+11M and above.

chalsall · 2012-02-21, 23:47

Quote:

Originally Posted by bcp19

Then again, why limit it to Anonymous? Isn't there currently a 'trust' level? If you had say 5 trust levels (anon = 5), Level 1 could get from lowest (x) to x+2M exp, level 2 from x+2M to x+4M, level 3 from x+4M to x+7M, level 4 from x+7M to x+11M and level 5 x+11M and above.

I don't disagree.

But, empricially, it seems there are only two trust levels with PrimeNet currently.

"Trusted" machines get "prefered" assignments.

Everyone else (including Anonymous machines) get anything above the prefered which are available; including those candidates at the "wave-back".

IMHO, PrimeNet needs some tender loving attention on its heuristics (and a bit of time on the table's indexes probably wouldn't hurt either)....

Dubslow · 2012-02-22, 00:07

Quote:

Originally Posted by cheesehead

My contention is still that unilaterally poaching someone else's assignment is never the best way to handle the situation.

At the very least, communicate with the current assignee and negotiate a cooperative resumption.

If that's not possible, contact George or other admin, explain the situation, and see what can be done without ambushing the current assignee with a poach.

It's not unilaterally poaching if you have a have-finished expo that was unreserved due to a bug in either P95 or PM. It is poaching, in one sense, but in many others it is not. It is not knowingly and willingly starting an assignment from scratch while also knowing someone else is in progress and might not get credit if you beat them. Therefore I would not call it poaching in the usual sense.

Communicating with the current assignee is in general not possible; I am sure that many people join GIMPS because of the minimum of interaction required to participate in the project.

Communicating with anyone, George/admin or not, is still a hassle, and falls under the "not-minimal-interaction". (Obviously I'm not talking about me personally, but in general.) It is far easier and simpler just to finish the work that had already been done so that I don't lose two months of work. No one will die.

Edit: Whoops, missed this page of posts.

As for PrimeNet, I think the main problems are as follows:

1) Windows
2) Internet Information Services
and most (or least) importantly,

3) There are like 3 people who have access to the server code, none of which have time to actually do anything about it. I don't think I've ever actually seen Scott around here. I propose that one or two more of us be given access to help maintain PrimeNet. (I would volunteer myself, except I don't think my HTML/PHP/SQL/etc else skills are anywhere near good enough.)

2012-02-21, 22:09	#408
garo Aug 2002 Termonfeckin, IE 2²×691 Posts	I can testify that there have been at least 4 occasions since I joined GIMPS when George manually released exponents that had been reserved for too long. I haven't seen one in at least a couple of years though. @Brian-E: Your supposition is incorrect. I for one had some exponents that I had been working on poached by aketilander while my main rig was down due to hardware issues. I sent him a PM but he said he was going to poach anyway. I can only hope that with more time on the project he is less eager to poach. Last fiddled with by garo on 2012-02-21 at 22:15

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Newer X64 build needed	Googulator	Msieve	73	2020-08-30 07:47
Performance of cuda-ecm on newer hardware?	fivemack	GMP-ECM	14	2015-02-12 20:10
Cause this don't belong in the milestone thread	bcp19	Data	30	2012-09-08 15:09
Newer msieves are slow on Core i7	mklasson	Msieve	9	2009-02-18 12:58
Use of large memory pages possible with newer linux kernels	Dresdenboy	Software	3	2003-12-08 14:47

2012-02-21, 22:19	#412
bcp19 Oct 2011 7×97 Posts	How difficult would it be, on assignments over 1 year old, to do something like add another column or two to the database, and store % complete and date that is not overwritten for say 30 days, to track progress? Thus today, it would store 2/21/12 and 14.6%. If the assignment completes, it's obviously no longer tracked, but after say 30 days, it could check progress and if less than a certain % cancel the assignment. I don't know if P95 has the programming to toss an assignment if flagged by the server, but it seems like something to consider adding.

2012-02-21, 23:03	#414
chalsall If I May "Chris Halsall" Sep 2002 Barbados 9782₁₀ Posts	Here's an idea (which has been suggested before)... Perhaps those who are "Anonymous" should be assigned work *well* ahead of the "waves".