OFFICIAL "SERVER PROBLEMS" THREAD - Page 262

PhilF · 2023-01-29, 20:38

Quote:

Originally Posted by James Heinrich

Thanks, fixed. The structure of the form was changed a couple months ago, I missed updating the check-all button.

I don't know how good we are at finding the next Mersenne Prime, but we're sure good at keeping you on your toes...

Mark Rose · 2023-01-29, 21:49

In the email the server sends if a computer is late contacting GIMPS, there is no space between "than" and the number of hours: e.g. "Your computer named foo is more than25 hours late reporting to the GIMPS server."

Prime95 · 2023-01-30, 01:14

Quote:

Originally Posted by Mark Rose

In the email the server sends if a computer is late contacting GIMPS, there is no space between "than" and the number of hours: e.g. "Your computer named foo is more than25 hours late reporting to the GIMPS server."

Fixed.

kriesel · 2023-01-30, 14:47

Server declares the absence of a reported proof is because of old software, even when the result is from a proof-capable software version. (Server is apparently ignoring the possibilities of user error such as failing to upload the proof file generated, or because of lengthy loss of connection to a network share, prime95 filling emergency memory and then losing necessary temporaries, or gpuowl having an error such as a checksum failure, such that it ceases to save needed temporaries to the local hard drive.)
Example: https://www.mersenne.org/report_expo...=1&swversion=1 gpuowl v6.11-380 is definitely proof-capable (as is -318 or above), but a combination of errors occurred such that past ~33%, temporaries were not saved. In this case it was not disk space issue, but some sort of execution issue. The last temporary saved in this case was at iteration 42506688, ~32.81%, so not enough to salvage a proof power of one or higher.

Code:

2023-01-22 12:42:43 asr2/radeonii1 129543839 OK 42400000  32.73%; 1073 us/it; ETA 1d 01:58; 4e66f55e3e1e01e0 (check 1.34s) 1 errors
2023-01-22 12:44:32 asr2/radeonii1 129543839 OK 42500000  32.81%; 1070 us/it; ETA 1d 01:52; ab53474db981f4e3 (check 1.33s) 1 errors
2023-01-22 12:46:23 asr2/radeonii1 GPU -> Host read #0 failed (check 8926b vs 6f95c37)
2023-01-22 12:46:23 asr2/radeonii1 GPU -> Host read #1 failed (check ffeaf200 vs 6f95c37)
2023-01-22 12:46:23 asr2/radeonii1 GPU -> Host read #2 failed (check ffeaf200 vs 6f95c37)
2023-01-22 12:46:23 asr2/radeonii1 Exiting because "GPU -> Host persistent read errors"
2023-01-22 12:46:23 asr2/radeonii1 Bye

Code:

2023-01-22 12:46:23 Note: not found 'config.txt'
2023-01-22 12:46:23 config: -device 3 -user kriesel -cpu asr2/radeonii3 -block 1000 -log 10000 -use NO_ASM -proof 10 -maxAlloc 15000 
2023-01-22 12:46:23 device 3, unique id ''
2023-01-22 12:46:23 asr2/radeonii3 129543839 FFT: 7M 1K:14:256 (17.65 bpw)
2023-01-22 12:46:23 asr2/radeonii3 Expected maximum carry32: 3F2A0000
2023-01-22 12:46:25 asr2/radeonii3 OpenCL args "-DEXP=129543839u -DWIDTH=1024u -DSMALL_HEIGHT=256u -DMIDDLE=14u -DPM1=0 -DAMDGPU=1 -DWEIGHT_STEP_MINUS_1=0x8.d0d2334fe6e4p-5 -DIWEIGHT_STEP_MINUS_1=-0xd.d2bfc8a870578p-6 -DNO_ASM=1  -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2023-01-22 12:46:29 asr2/radeonii3 OpenCL compilation in 4.43 s
2023-01-22 12:46:33 asr2/radeonii3 129543839 OK 42500000 loaded: blockSize 1000, ab53474db981f4e3
2023-01-22 12:46:33 asr2/radeonii3 validating proof residues for power 10
2023-01-22 12:47:48 asr2/radeonii3 checksum b3880f0d (expected 3e6bd075) in '.\129543839\proof\37193352'
2023-01-22 12:47:48 asr2/radeonii3 validating proof residues for power 9
2023-01-22 12:48:26 asr2/radeonii3 checksum b3880f0d (expected 3e6bd075) in '.\129543839\proof\37193352'
2023-01-22 12:48:26 asr2/radeonii3 validating proof residues for power 8
2023-01-22 12:48:26 asr2/radeonii3 Can't open '.\129543839\proof\506031' (mode 'rb')
2023-01-22 12:48:26 asr2/radeonii3 validating proof residues for power 7
2023-01-22 12:48:26 asr2/radeonii3 Can't open '.\129543839\proof\1012062' (mode 'rb')
2023-01-22 12:48:26 asr2/radeonii3 validating proof residues for power 6
2023-01-22 12:48:26 asr2/radeonii3 Can't open '.\129543839\proof\2024123' (mode 'rb')
2023-01-22 12:48:26 asr2/radeonii3 Proof disabled because of missing checkpoints
2023-01-22 12:48:33 asr2/radeonii3 129543839 OK 42502000  32.81%; 2186 us/it; ETA 2d 04:52; ab64d70bca6339fa (check 2.51s) 1 errors
2023-01-22 12:48:52 asr2/radeonii3 129543839 OK 42510000  32.82%; 2148 us/it; ETA 2d 03:56; 93083a22b97ed44b (check 2.49s) 1 errors
2023-01-22 12:49:16 asr2/radeonii3 129543839 OK 42520000  32.82%; 2152 us/it; ETA 2d 04:02; 679dd0be37d47507 (check 2.55s) 1 errors
...

(In V6.11-380 and similar, there is a slight iteration count numbering difference versus proof power, such that a power 10 run saves iteration 506032 in the example run, while validation of power 8 seeks 506031, so fallback to 8 on a power 10 run fails. IIRC that was corrected in v7.x)

Second example, prime95 v30.8b14: 480000107
Occasionally in prime.log:

Code:

DOSerrno: 53
Cannot open PRP proof interim residues file: \\Ostrich\prp-proof-cache\p480000107.residues
Errno: 2, No such file or directory

James Heinrich · 2023-01-30, 17:58

Quote:

Originally Posted by kriesel

Server declares the absence of a reported proof is because of old software, even when the result is from a proof-capable software version. (Server is apparently ignoring the possibilities of user error such as failing to upload the proof file generated, or because of lengthy loss of connection to a network share, prime95 filling emergency memory and then losing necessary temporaries, or gpuowl having an error such as a checksum failure, such that it ceases to save needed temporaries to the local hard drive.)
Example: 129543839 gpuowl v6.11-380 is definitely proof-capable
Second example, prime95 v30.8b14: 480000107

The criteria for "PRP without proof (old software)" is whether or not the JSON result contains the "proof" section. This message does not differentiate between "PRP software didn't generate a proof because it doesn't know how to" vs "PRP software didn't generate a proof because of some error on the user's system". What it is differentiating between is "PRP without proof because it wasn't uploaded" vs "PRP without proof because it wasn't generated".

kriesel · 2023-01-30, 18:06

So, perhaps it should say proof not generated, instead of claiming the sometimes-wrong "old software". Or better yet distinguish between them, but that would require more conditionals, on software-specific version thresholds.
There have been a tremendous number of improvements recently in server web page outputs and options. This is one more opportunity to better inform the users. As it stands, it is sometimes MISinforming the users.

James Heinrich · 2023-01-30, 18:11

Wording changed.

S485122 · 2023-02-02, 13:21

Quote:

Originally Posted by S485122

One the https://www.mersenne.org/primenet/ "PrimeNet Statistics" page, in the "Exponent Status Distribution" part, the 10M range still has one verified exponent too many, which, by the design of the page gives a negative "Untested" count. (As stated previously in this thread that last count is computed, not tallied).
Probably a wrongly counted PRP or CERT, perhaps about a cofactor.

Now there are 6 ranges with one or more "Verified" exponents too many :

2M -1
8M -1
9M -1
10M -1
14M -6

Andrew Usher · 2023-02-02, 13:47

I count 5 there, not 6, but the problem is evidently some exponents being classified as both 'factored' and 'verified', which are supposed to be mutually exclusive.

For another weird server mistake, look at M111597749, in the milestone list for a long time in what appears to be an abandoned LLDC - but it's not. The LL assignment was completed and returned 2 years ago, but never expired due to being wrongly (no previous test is recorded) classified as a DC. I imagine someone has to submit an actual DC to expire that one.

S485122 · 2023-02-04, 10:52

I believe I have identified the problem, looking at 14M PRP Results Page it is easy to verify that 6 of the "Verified PRP test results" should be in the "Factored PRP test results" section".

The wrongly classified PRP tests concern the following exponents in the 5 (indeed) ranges :

M2014997
M8833439
M9999991
M10000019
M14357429
M14357459
M14357489
M14357507
M14357527
M14357579

They were perhaps malformed attempts at cofactor PRP tests on exponents with factors ? But the result seems to be that the exponents status changed from factored to factored AND verified, and thus being counted twice.

Andrew Usher · 2023-02-04, 13:12

Yes, it appears that PRP tests without known factors were conducted and the server misinterprets these. They should be ignored as redundant - as an LL test would be - since the exponent is already factored.

2023-01-30, 18:06	#2877
kriesel "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 1CCB₁₆ Posts	So, perhaps it should say proof not generated, instead of claiming the sometimes-wrong "old software". Or better yet distinguish between them, but that would require more conditionals, on software-specific version thresholds. There have been a tremendous number of improvements recently in server web page outputs and options. This is one more opportunity to better inform the users. As it stands, it is sometimes MISinforming the users. Last fiddled with by kriesel on 2023-01-30 at 18:11

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Official "Faits erronés dans de belles-lettres" thread	ewmayer	Lounge	39	2015-05-19 01:08
Official "all-Greek-to-me Fiction Literature and Cinema" Thread	ewmayer	Science & Technology	41	2014-04-16 11:54
Official "Lasciate ogne speranza" whinge-thread	cheesehead	Soap Box	56	2013-06-29 01:42
Official "Ernst is a deceiving bully and George is a meanie" thread	cheesehead	Soap Box	61	2013-06-11 04:30
Official "String copy Statement Considered Harmful" thread	Dubslow	Programming	19	2012-05-31 17:49

2023-01-29, 21:49	#2873
Mark Rose "/X\(‘-‘)/X\" Jan 2013 6042₈ Posts	In the email the server sends if a computer is late contacting GIMPS, there is no space between "than" and the number of hours: e.g. "Your computer named foo is more than25 hours late reporting to the GIMPS server."

2023-01-30, 18:11	#2878
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 2·3·683 Posts	Wording changed.

2023-02-02, 13:47	#2880
Andrew Usher Dec 2022 2²×3×19 Posts	I count 5 there, not 6, but the problem is evidently some exponents being classified as both 'factored' and 'verified', which are supposed to be mutually exclusive. For another weird server mistake, look at M111597749, in the milestone list for a long time in what appears to be an abandoned LLDC - but it's not. The LL assignment was completed and returned 2 years ago, but never expired due to being wrongly (no previous test is recorded) classified as a DC. I imagine someone has to submit an actual DC to expire that one.

2023-02-04, 10:52	#2881
S485122 "Jacob" Sep 2006 Brussels, Belgium 1,907 Posts	I believe I have identified the problem, looking at 14M PRP Results Page it is easy to verify that 6 of the "Verified PRP test results" should be in the "Factored PRP test results" section". The wrongly classified PRP tests concern the following exponents in the 5 (indeed) ranges : M2014997 M8833439 M9999991 M10000019 M14357429 M14357459 M14357489 M14357507 M14357527 M14357579 They were perhaps malformed attempts at cofactor PRP tests on exponents with factors ? But the result seems to be that the exponents status changed from factored to factored AND verified, and thus being counted twice.

2023-02-04, 13:12	#2882
Andrew Usher Dec 2022 2²·3·19 Posts	Yes, it appears that PRP tests without known factors were conducted and the server misinterprets these. They should be ignored as redundant - as an LL test would be - since the exponent is already factored.