Poaching (please don't!)

[kriesel]

Maybe folks whose assignments get poached would find it useful to have a place to vent. Here it is. Halfway done or more, on a properly assigned exponent that figures in the next minor milestone, and discover someone else reported a result for it, with no indication they had the assignment? That seems rude.

Okay, I'll go first.

https://www.mersenne.org/report_expo...exp_hi=&full=1
assigned to me 2/27/2020

What's up with poaching a new assignment there, WR, poach result at only 3 days old assignment? Or did yours expire and get issued to me while your hardware soldiered on? It was / is going full tilt on one of my faster GPUs. GPU apps do not report progress to the PrimeNet server along the way, no matter how fast they're progressing, or how far they've gotten, because there's no software to do the PrimeNet API for gpu applications. And I've asked for the manual submission page to be extended to accept manually input progress lines, but it hasn't been implemented. So initiating a poach could be a setup to be running an imminent double-check instead. A really fast gpu can knock out a wavefront primality test in a day, so there's not much need for progress reporting there. It can take a few days for an exponent to get to the front of the worktodo queue. That all fits within a week. Not allowing even 4 days for me to complete a new assignment is pretty extreme. I get assigned one of these stragglers for the milestone very rarely. I was looking forward to making a difference for the next milestone.

I tried to PM or email WR, but I found no contact info. So there it is.

[Runtime Error]

Yeah, I purposely don't take those Cat 0 assignments anymore after getting poached by Curtis Cooper for the third time. I understand he just lets it run even after expiration & he's not purposefully poaching, but it still hurts.

[kriesel]

Originally Posted by Runtime Error

Yeah, I purposely don't take those Cat 0 assignments anymore after getting poached by Curtis Cooper for the third time. I understand he just lets it run even after expiration & he's not purposefully poaching, but it still hurts.

It would also be good if he (well, everyone) would switch from LL to PRP default for run of the mill future assignments.

[Uncwilly]

Originally Posted by kriesel

It would also be good if he (well, everyone) would switch from LL to PRP default for run of the mill future assignments.

CC has hundreds of deployed machines across the university. He does not have access to them all. Most are just set-up when they are new and that is it. If he is setting the new ones to do PRP, that is all that we can hope for.

[axn]

Originally Posted by Uncwilly

CC has hundreds of deployed machines across the university. He does not have access to them all. Most are just set-up when they are new and that is it. If he is setting the new ones to do PRP, that is all that we can hope for.

I do believe he has some form of capability to manage all the computers (at least as far as P95 goes). Has anyone written to him? Perhaps George can (officially). It is worth a shot. And maybe give a canned report which lists all the machines/tests with a GEC error count > 0 as well?

[retina]

Originally Posted by Runtime Error

Yeah, I purposely don't take those Cat 0 assignments anymore after getting poached by Curtis Cooper for the third time. I understand he just lets it run even after expiration & he's not purposefully poaching, but it still hurts.

Maybe the expiration criteria need adjusting.

Instead of blaming the user, perhaps the assignment code need adjusting to detect potential conflicts and avoid assigning it to another user.

[Runtime Error]

Originally Posted by retina

Maybe the expiration criteria need adjusting.

Instead of blaming the user, perhaps the assignment code need adjusting to detect potential conflicts and avoid assigning it to another user.

Oh yeah I definitely don't blame him. Also, I try to do 10% of my GHz/days as DC so it's not even that. But it still feels bad getting poached. Psychology is weird.

Originally Posted by Uncwilly

CC has hundreds of deployed machines across the university. He does not have access to them all. Most are just set-up when they are new and that is it. If he is setting the new ones to do PRP, that is all that we can hope for.

Don't they have to be linked to his account in order to get credit for the results? It can be changed on the computer tab under work preference. It's a lot of computers, but a student could do it (and I did more menial work as an RA myself lol).

[kriesel]

Originally Posted by Runtime Error

Oh yeah I definitely don't blame him. Also, I try to do 10% of my GHz/days as DC so it's not even that. But it still feels bad getting poached. Psychology is weird.

Don't they have to be linked to his account in order to get credit for the results? It can be changed on the computer tab under work preference. It's a lot of computers, but a student could do it (and I did more menial work as an RA myself lol).

10% DC, 90% first-time would mean accepting a DC delay of about 13 years.
(1/9)^(1/2.1) x 93M = 32.66M. First-testing now is occurring at 93M; 32.66M was getting first-tested mostly in 2006-2007.
It may be a lot of computers, but they could probably be churned out at some hundreds per hour per person with a good internet connection.

[Runtime Error]

Originally Posted by kriesel

10% DC, 90% first-time would mean accepting a DC delay of about 13 years.

Sure. I only mentioned that I do a little DC to highlight the fact that it still *feels* bad getting poached on a first time test even when I also choose to do some DC. Perhaps a more rational individual wouldn't care as much since it still counts as "finishing" checking an exponent with DC. I guess I'm not in Q.

On an unrelated note, I just returned to the US (forgive the delayed response) from the middle east. There were people on my flight from Europe via connections. There was absolutely zero corona virus screening in the US airport, but there was screening when we landed over there. Yikes!

[Rodrigo]

Originally Posted by retina

Instead of blaming the user, perhaps the assignment code need adjusting to detect potential conflicts and avoid assigning it to another user.

A change like that would be highly welcome. It happens to me several times a year that I'll submit a GPU TF report and have some (sometimes all) of the data rejected because somebody else TF'd the exponents after I received the assignment. Happened again this morning, matter of fact. Fortunately, in this case I caught it before spending any processor cycles or kWh on them.

[kladner]

Originally Posted by retina

Maybe the expiration criteria need adjusting.

Instead of blaming the user, perhaps the assignment code need adjusting to detect potential conflicts and avoid assigning it to another user.

Bravo! While there are some users who just work an area without regard to others, my impression of CC says that he would not intentionally poach. When running such a large, (and productive) farm there has to be some reliance on default settings. To monitor ~900-100 systems might have to rely on trouble notifications (emails?) from individual systems. There are articles and interviews online with Curtis Cooper. Worth looking up.