Register FAQ Search Today's Posts Mark Forums Read

2019-05-22, 20:41   #1717
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

512410 Posts

Quote:
 Originally Posted by GP2 the worktodo file and the All CPUs section of https://www.mersenne.org/workload/ had conflicting information for the residue type for the same assignment. So the All CPUs section should instead accurately reflect the actual residue type for the assignment.
Exactly!
For the same single assignment, I looked 3 places.
The assignments page was internally consistent, but did not match what the prime95 instance had in its worktodo file and so is presumably doing. The two PRP= lines given below are the same aid/assignment, and should match exactly, but differ in residue type.

Code:
This is a PRP DC delivered by the Primenet API;
PRP=(aid),1,2,79335979,-1,75,0,3,1 in the worktodo file,
PRP=(aid),1,2,79335979,-1,75,0,3,4 in the All cpus and system-specific listings on my assignments page
Checking again today:
Code:
PRP=EE...A0,1,2,79335979,-1,75,0,3,4 in All CPUs section of assignments page
PRP=EE...A0,1,2,79335979,-1,75,0,3,4 in cpu-specific (Parrot) section of assignments page
PRP=EE...A0,1,2,79335979,-1,75,0,3,1 in worktodo.txt on Parrot prime95 instance
Same everything, for same assignment, except residue type 1 in the worktodo is not a match to 4 in the assignment page.
It would be useful if prime95 indicated in its worker window title bar whether it thinks it's doing a first test or doublecheck, or what residue type, and indicated in its log files what residue type etc it got assigned. It only indicates when assigned, exponent, AID, and PRP, in prime.log.

2019-05-22, 22:21   #1718
James Heinrich

"James Heinrich"
May 2004
ex-Northern Ontario

3,361 Posts

Quote:
 Originally Posted by kriesel Why not have the page provide defaults, that correspond to the category selected
It does now do that.

 2019-05-23, 03:53 #1719 Rodrigo     Jun 2010 Pennsylvania 2×467 Posts Not counting the "what makes sense" choices, there are 12 permutations of choices on the Manual GPU Assignments page (3 for preferred work range * 4 for work preference). I just tried all of them; the only combinations of choices that yielded anything other than an Error code 40, are TF for 100M with any of the four work preference choices -- and those all gave assignments in the 332M range. The default ranges did show up in the "Optional exponent range field," but the 8 permutations involving TF for DCs and LLs all returned that Error code 40 for me.
2019-05-23, 04:45   #1720
Serpentine Vermin Jar

Jul 2014

37·89 Posts

Quote:
 Originally Posted by GP2 The ,75,0,3, has a zero in the middle, which is the number of tests saved. This parameter only makes sense for P−1 assignments, and is just a placeholder here, so it can always be set to zero. Also, I'm not sure if I'm correctly interpreting what kriesel wrote, but it seems like the worktodo file and the All CPUs section of https://www.mersenne.org/workload/ had conflicting information for the residue type for the same assignment. So the All CPUs section should instead accurately reflect the actual residue type for the assignment.
Whoops, I forgot about that 0 in there showing how many tests would be saved.

For what it's worth, the /workload page will *always* show 0 there. On the other hand, when the client gets the assignment from the server that first time, it should be correct.

What Kriesel said was interesting, because if I look at that example correctly, the one that the "/workload" page shows should be the correct one since it matches the most recent previous test.

Maybe the auto-assignment got confused in that case because there was the earlier test with different base/type values. If so, that seems like a bug in the assignment generator where it didn't pick the most recent PRP test to match? Not sure.

Sometimes it's hard for me to know what is by design and what's by accident.

2019-05-23, 16:02   #1721
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

120048 Posts

Quote:
 Originally Posted by James Heinrich It does now do that.
Tried it out this morning, got for first time LL TF, defaults, lowest exponents preference, lower limit surprisingly low at 83+M (well below 91+million where P-1 is now being assigned, or even 85M where primality tests are being assigned), and an upper limit of 93 million, which resulted in error 40 no assignments available. Bumped the upper limit to 95 million, got assignments 9305xxxx. Maybe the ranges selected are off by one in some table?

It would be nice if clicking "get more assignments" displayed a page with the previous choices made in that browser /page session, rather than resetting to default load on everything.

If I want ALL the same choices, including number of assignments, just reload page works to get another batch of assignments,
but if I want to change just one field, it's "get more assignments" and then set everything again.

Last fiddled with by kriesel on 2019-05-23 at 16:13

2019-05-23, 17:17   #1722
James Heinrich

"James Heinrich"
May 2004
ex-Northern Ontario

3,361 Posts

Quote:
 Originally Posted by kriesel ... upper limit of 93 million, which resulted in error 40 no assignments available. Bumped the upper limit to 95 million, got assignments 9305xxxx. Maybe the ranges selected are off by one in some table?
Not really, but the upper limit is dynamically chosen based on the "active range" plus a little bit. I've bumped the upper limit by +1M so it should have a better chance of finding available work now.

Quote:
 Originally Posted by kriesel It would be nice if clicking "get more assignments" displayed a page with the previous choices made in that browser /page session, rather than resetting to default load on everything.
Done.

2019-05-23, 17:58   #1723
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

22×3×7×61 Posts

Quote:
 Originally Posted by Madpoo Sometimes it's hard for me to know what is by design and what's by accident.
There's another use for design and documentation.

 2019-05-23, 18:04 #1724 Rodrigo     Jun 2010 Pennsylvania 16468 Posts I just went back into the Manual GPU Assignments page to try things out. This time, none of the 12 combinations of choices yields anything. Unlike last night, there are no Error 40's; instead, however, the favicon in the browser tab turns into a circle, spinning endlessly with no result. Logged in on different browsers on different computers, same outcome. Internet connectivity is otherwise fine. Just prior to posting this, the circle stopped spinning, the favicon returned to the Mersenne.org standard, and the results pane went blank. (See attachment.) Attached Thumbnails
2019-05-23, 18:59   #1725
James Heinrich

"James Heinrich"
May 2004
ex-Northern Ontario

3,361 Posts

Quote:
 Originally Posted by Rodrigo spinning endlessly with no result
I couldn't replicate that, but I did fix a small error that might have caused it for you. Please try again?

 2019-05-23, 19:24 #1726 Rodrigo     Jun 2010 Pennsylvania 2·467 Posts The change worked nicely! Now all the LL and 100M combinations work. Only the DCs continue to lead to an Error code 40 here. Thank you!
2019-05-23, 20:05   #1727
James Heinrich

"James Heinrich"
May 2004
ex-Northern Ontario

3,361 Posts

Quote:
 Originally Posted by Rodrigo Now all the LL and 100M combinations work. Only the DCs continue to lead to an Error code 40 here.
According to the data I'm seeing, DC's should be around ~55M right now, but the function that fetches available assignments isn't finding anything below 94M, which isn't right. I'll need George or Aaron to look into that.

 Similar Threads Thread Thread Starter Forum Replies Last Post ewmayer Lounge 39 2015-05-19 01:08 ewmayer Science & Technology 41 2014-04-16 11:54 cheesehead Soap Box 56 2013-06-29 01:42 cheesehead Soap Box 61 2013-06-11 04:30 Dubslow Programming 19 2012-05-31 17:49

All times are UTC. The time now is 13:54.

Fri May 14 13:54:40 UTC 2021 up 36 days, 8:35, 0 users, load averages: 1.64, 1.78, 1.81