Prime95 v30.3 - Page 17

kruoli · 2020-08-29, 09:15

On a lot of machines and fore some versions now (I cannot recall if it ever was different), Prime95 never was able to restore the maximized status of it's (only) subwindow.

Prime95 · 2020-08-30, 19:49

Build 4 is available. Go to the first post in this thread to download.

This version features an overhaul of the CERT assignment strategy. Looking at the (Ben Delo) data it seems the default settings did not assign CERTs in roughly the same proportion as results. produced. Ben was doing about 2 CERTs for every 3 results.

This was due to prime95 requesting making the same number of get cert assignment calls each day regardless of CPU power. That is, a slow dual core CPU would get, on average, the same number of CERT assignments each day as a 24-core AWS server.

Build 3 would check for an assignment at startup, every 6 hours, after each result reported, and when sending new expected completion dates.

Build 4 will check for an assignment at startup and every N hours where N depends on the number of CPU cores:

Code:

    20+,     N=3
    12-19,   N=4
    7-11,    N=6
    3-6,     N=8
    2        N=12
    1        N=24

This should lead to a more equitable distribution.

Other changes:
1) Getting PRP-CF CERTs is disabled by default. To turn it on, put CertMinExponent=0 in local.txt. The default is 50 million. I've set my computers to accept PRP-CF CERT assignments which should easily handle the load.
2) If you are assigned a PRP-CF CERT, prime95 will get up to 5 at a time. They are quick and this might minimize interruptions.
3) If you set CERT CPU percentage to 50 or more in the advanced resources dialog, then you will be considered a "CERT volunteer". Prime95 will check for CERT assignments every half hour even if you already have a CERT assignment, and will get multiple CERT assignments at a time.
4) You will not get CERT assignments on exponents over 200 million unless you have configured prime95 with an average of 4 or more threads per worker.

There are local.txt options to override the get assignment frequency and the quantity that can be queued up at one time. If there is a good reason to make these options public, I'll add them to undoc.txt.

keisentraut · 2020-08-31, 10:25

I think the server upload is congested. I run into timeouts, using an ISP in Germany. Speed of downloading the new mprime build was very slow, too.

Please note that you easily might create an unintended DDoS: if everyone's download fails, and all retry, it will only get worse and worse.

Quote:

[Work thread Aug 31 12:13] Starting certification of M99838889 using FMA3 FFT length 5376K, Pass1=896, Pass2=6K, clm=1, 2 threads
[Comm thread Aug 31 12:13] URL: http://www.mersenne.org/proof_get_da...344F1124AE8E94
[Comm thread Aug 31 12:16] CURL library error: Operation timed out after 180000 milliseconds with 9185022 out of 12479894 bytes received
[Comm thread Aug 31 12:16] CURL library error: Operation timed out after 180000 milliseconds with 9185022 out of 12479894 bytes received
[Work thread Aug 31 12:16] Error getting CERT starting value.
[Work thread Aug 31 12:16] Will retry certification later.
[Work thread Aug 31 12:16] Aborting processing of this work unit -- will try again later.

keisentraut · 2020-08-31, 10:56

Another issue: Here is an example of Ben Delo having certified his own result: M99841073. I do not understand the math enough, but would this enable Ben Delo to "cheat"? Maybe the server should not assign CERT jobs to the user did the PRP (or "Anonymous").
(PS: Of course I trust Ben Delo not to "cheat". But someone else might.)

moebius · 2020-08-31, 11:23

Quote:

Originally Posted by keisentraut

Another issue: Here is an example of Ben Delo having certified his own result

yes, we will soon be obsolete for Delo

LaurV · 2020-08-31, 11:53

Quote:

Originally Posted by keisentraut

... having certified his own result...

There is no problem with people certifying their own results. You can't "cheat". Well, not in an easy/reasonable/efficient way, at least. Trying to falsify a cert, you will need to invest much more resources than running the cert, and your chances of success are extremely slim.

jwnutter · 2020-08-31, 13:16

Quote:

Originally Posted by Prime95

Set MaxProofgenWaits=1 in prime.txt

The default is to try proof generation for 2 days before giving up (in case the file is on an NFS mount that is offline).
It is on my to-do list to tailor the wait time to the error. A file open error might wait two days, while your MD5 error is
unlikely to get better - maybe try just twice.

I also received the error: "MD5 error reading PRP proof interim residues file." I shut down P95, set MaxProofgenWaits=1 in Prime.txt, restarted, and all seems to be well now. I'm not 100% certain, but it appears P95 is working on the proof file now. Interestingly, I have two files titled prime.txt in the same P95 folder - I'm not sure how or why. It appears that both files are being updated with different information.

Currently running v30.3, build 3.

Code:

[Aug 31 06:32] Iteration: 83780000 / 83813888 [99.95%], ms/iter:  3.790, ETA: 00:02:08
[Aug 31 06:33] Iteration: 83790000 / 83813888 [99.97%], ms/iter:  3.792, ETA: 00:01:30
[Aug 31 06:34] Iteration: 83800000 / 83813888 [99.98%], ms/iter:  3.805, ETA: 00:00:52
[Aug 31 06:34] Iteration: 83810000 / 83813888 [99.99%], ms/iter:  3.790, ETA: 00:00:14
[Aug 31 06:34] Gerbicz error check passed at iteration 83813604.
[Aug 31 06:35] Gerbicz error check passed at iteration 83813860.
[Aug 31 06:35] Generating proof for M83813623.  Proof power = 9, Hash length = 64
[Aug 31 06:35] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 06:35] hash0 = 6EB4CA99EB29F88D
[Aug 31 06:35] hash1 = FEF3ECA63085725C
[Aug 31 06:35] hash2 = 94413FE21F1716F1
[Aug 31 06:35] hash3 = F9034C494D5A28C8
[Aug 31 06:35] hash4 = 75DBBC01ED030E0E
[Aug 31 06:36] hash5 = F585034046BE90F7
[Aug 31 06:37] hash6 = 3EBCB6D5E1805F1C
[Aug 31 06:38] hash7 = 7E571EC7FA97D885
[Aug 31 06:38] MD5 error reading PRP proof interim residues file.
[Aug 31 06:38] Waiting 5 minutes to try proof generation again.
[Aug 31 06:38] Waiting five minutes before restarting.
[Aug 31 06:44] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 06:44] hash0 = 6EB4CA99EB29F88D
[Aug 31 06:44] hash1 = FEF3ECA63085725C
[Aug 31 06:44] hash2 = 94413FE21F1716F1
[Aug 31 06:44] hash3 = F9034C494D5A28C8
[Aug 31 06:44] hash4 = 75DBBC01ED030E0E
[Aug 31 06:45] hash5 = F585034046BE90F7
[Aug 31 06:46] hash6 = 3EBCB6D5E1805F1C
[Aug 31 06:48] hash7 = 7E571EC7FA97D885
[Aug 31 06:48] MD5 error reading PRP proof interim residues file.
[Aug 31 06:48] Waiting 5 minutes to try proof generation again.
[Aug 31 06:48] Waiting five minutes before restarting.
[Aug 31 06:53] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 06:53] hash0 = 6EB4CA99EB29F88D
[Aug 31 06:53] hash1 = FEF3ECA63085725C
[Aug 31 06:53] hash2 = 94413FE21F1716F1
[Aug 31 06:53] hash3 = F9034C494D5A28C8
[Aug 31 06:54] hash4 = 75DBBC01ED030E0E
[Aug 31 06:54] hash5 = F585034046BE90F7
[Aug 31 06:55] hash6 = 3EBCB6D5E1805F1C
[Aug 31 06:57] hash7 = 7E571EC7FA97D885
[Aug 31 06:57] MD5 error reading PRP proof interim residues file.
[Aug 31 06:57] Waiting 5 minutes to try proof generation again.
[Aug 31 06:57] Waiting five minutes before restarting.
[Aug 31 07:02] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 07:02] hash0 = 6EB4CA99EB29F88D
[Aug 31 07:02] hash1 = FEF3ECA63085725C
[Aug 31 07:02] hash2 = 94413FE21F1716F1
[Aug 31 07:02] hash3 = F9034C494D5A28C8
[Aug 31 07:03] hash4 = 75DBBC01ED030E0E
[Aug 31 07:03] hash5 = F585034046BE90F7
[Aug 31 07:04] hash6 = 3EBCB6D5E1805F1C
[Aug 31 07:06] hash7 = 7E571EC7FA97D885
[Aug 31 07:06] MD5 error reading PRP proof interim residues file.
[Aug 31 07:06] Waiting 5 minutes to try proof generation again.
[Aug 31 07:06] Waiting five minutes before restarting.
[Aug 31 07:11] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 07:11] hash0 = 6EB4CA99EB29F88D
[Aug 31 07:11] hash1 = FEF3ECA63085725C
[Aug 31 07:12] hash2 = 94413FE21F1716F1
[Aug 31 07:12] hash3 = F9034C494D5A28C8
[Aug 31 07:12] hash4 = 75DBBC01ED030E0E
[Aug 31 07:12] hash5 = F585034046BE90F7
[Aug 31 07:13] hash6 = 3EBCB6D5E1805F1C
[Aug 31 07:15] hash7 = 7E571EC7FA97D885
[Aug 31 07:15] MD5 error reading PRP proof interim residues file.
[Aug 31 07:15] Waiting 5 minutes to try proof generation again.
[Aug 31 07:15] Waiting five minutes before restarting.
[Aug 31 07:21] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 07:21] hash0 = 6EB4CA99EB29F88D
[Aug 31 07:21] hash1 = FEF3ECA63085725C
[Aug 31 07:21] hash2 = 94413FE21F1716F1
[Aug 31 07:21] hash3 = F9034C494D5A28C8
[Aug 31 07:21] hash4 = 75DBBC01ED030E0E
[Aug 31 07:22] hash5 = F585034046BE90F7
[Aug 31 07:23] hash6 = 3EBCB6D5E1805F1C
[Aug 31 07:24] hash7 = 7E571EC7FA97D885
[Aug 31 07:24] MD5 error reading PRP proof interim residues file.
[Aug 31 07:24] Waiting 5 minutes to try proof generation again.
[Aug 31 07:24] Waiting five minutes before restarting.
[Aug 31 07:30] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 07:30] hash0 = 6EB4CA99EB29F88D
[Aug 31 07:30] hash1 = FEF3ECA63085725C
[Aug 31 07:30] hash2 = 94413FE21F1716F1
[Aug 31 07:30] hash3 = F9034C494D5A28C8
[Aug 31 07:30] hash4 = 75DBBC01ED030E0E
[Aug 31 07:31] hash5 = F585034046BE90F7
[Aug 31 07:32] hash6 = 3EBCB6D5E1805F1C
[Aug 31 07:34] hash7 = 7E571EC7FA97D885
[Aug 31 07:34] MD5 error reading PRP proof interim residues file.
[Aug 31 07:34] Waiting 5 minutes to try proof generation again.
[Aug 31 07:34] Waiting five minutes before restarting.
[Aug 31 07:39] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 07:39] hash0 = 6EB4CA99EB29F88D
[Aug 31 07:39] hash1 = FEF3ECA63085725C
[Aug 31 07:39] hash2 = 94413FE21F1716F1
[Aug 31 07:39] hash3 = F9034C494D5A28C8
[Aug 31 07:39] hash4 = 75DBBC01ED030E0E
[Aug 31 07:40] hash5 = F585034046BE90F7
[Aug 31 07:41] hash6 = 3EBCB6D5E1805F1C
[Aug 31 07:43] hash7 = 7E571EC7FA97D885
[Aug 31 07:43] MD5 error reading PRP proof interim residues file.
[Aug 31 07:43] Waiting 5 minutes to try proof generation again.
[Aug 31 07:43] Waiting five minutes before restarting.
[Aug 31 07:48] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 07:48] hash0 = 6EB4CA99EB29F88D
[Aug 31 07:48] hash1 = FEF3ECA63085725C
[Aug 31 07:48] hash2 = 94413FE21F1716F1
[Aug 31 07:48] hash3 = F9034C494D5A28C8
[Aug 31 07:49] hash4 = 75DBBC01ED030E0E
[Aug 31 07:49] hash5 = F585034046BE90F7
[Aug 31 07:50] hash6 = 3EBCB6D5E1805F1C
[Aug 31 07:52] hash7 = 7E571EC7FA97D885
[Aug 31 07:52] MD5 error reading PRP proof interim residues file.
[Aug 31 07:52] Waiting 5 minutes to try proof generation again.
[Aug 31 07:52] Waiting five minutes before restarting.
[Aug 31 07:57] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 07:57] hash0 = 6EB4CA99EB29F88D
[Aug 31 07:57] hash1 = FEF3ECA63085725C
[Aug 31 07:58] hash2 = 94413FE21F1716F1
[Aug 31 07:58] hash3 = F9034C494D5A28C8
[Aug 31 07:58] hash4 = 75DBBC01ED030E0E
[Aug 31 07:58] hash5 = F585034046BE90F7
[Aug 31 07:59] hash6 = 3EBCB6D5E1805F1C
[Aug 31 08:01] hash7 = 7E571EC7FA97D885
[Aug 31 08:01] MD5 error reading PRP proof interim residues file.
[Aug 31 08:01] Waiting 5 minutes to try proof generation again.
[Aug 31 08:01] Waiting five minutes before restarting.
[Aug 31 08:01] Worker stopped.
[Aug 31 08:04] Worker starting
[Aug 31 08:04] Setting affinity to run worker on CPU core #1
[Aug 31 08:04] Setting affinity to run helper thread 1 on CPU core #2
[Aug 31 08:04] Setting affinity to run helper thread 4 on CPU core #5
[Aug 31 08:04] Setting affinity to run helper thread 2 on CPU core #3
[Aug 31 08:04] Setting affinity to run helper thread 5 on CPU core #6
[Aug 31 08:04] Setting affinity to run helper thread 6 on CPU core #7
[Aug 31 08:04] Setting affinity to run helper thread 3 on CPU core #4
[Aug 31 08:04] Setting affinity to run helper thread 7 on CPU core #8
[Aug 31 08:04] Resuming Gerbicz error-checking PRP test of M83813623 using FMA3 FFT length 4480K, Pass1=896, Pass2=5K, clm=1, 8 threads
[Aug 31 08:04] PRP proof using power=9 and 64-bit hash size.
[Aug 31 08:04] Proof requires 5.4GB of temporary disk space and uploading a 105MB proof file.
[Aug 31 08:04] Iteration: 83389522 / 83813888 [99.49%].
[Aug 31 08:04] Iteration: 83390000 / 83813888 [99.49%], ms/iter:  3.796, ETA: 00:26:49
[Aug 31 08:04] Iteration: 83400000 / 83813888 [99.50%], ms/iter:  3.798, ETA: 00:26:12

retina · 2020-08-31, 13:22

Quote:

Originally Posted by jwnutter

Interestingly, I have two files titled prime.txt in the same P95 folder ...

They have different icons, so one of them is probably a word file *.doc or something.

Great to see MS are still insisting upon hiding extensions and confusing everybody with crap like that! I would suggest that for everyone to untick that option and show the full file names.

Viliam Furik · 2020-08-31, 13:26

Quote:

Originally Posted by retina

They have different icons, so one of them is probably a word file *.doc or something.

Great to see MS are still insisting upon hiding extensions and confusing everybody with crap like that! I would suggest that for everyone to untick that option and show the full file names.

One of them is prime.log, the other is prime.txt. I think the one with a blue square in the icon is the .log

jwnutter · 2020-08-31, 13:37

Received another MD5 Error. Maybe I should upgrade to build 4 and try again.

Code:

[Aug 31 08:30] Iteration: 83800000 / 83813888 [99.98%], ms/iter:  3.741, ETA: 00:00:51
[Aug 31 08:30] Iteration: 83810000 / 83813888 [99.99%], ms/iter:  3.727, ETA: 00:00:14
[Aug 31 08:31] Gerbicz error check passed at iteration 83813604.
[Aug 31 08:31] Gerbicz error check passed at iteration 83813860.
[Aug 31 08:31] Generating proof for M83813623.  Proof power = 9, Hash length = 64
[Aug 31 08:31] Root hash = 0D5CCB0B72A6FEB1AF3678FE396B0B3F9CA1DC9B05AF3A7A74D720E2A768F976
[Aug 31 08:31] hash0 = 6EB4CA99EB29F88D
[Aug 31 08:31] hash1 = FEF3ECA63085725C
[Aug 31 08:31] hash2 = 94413FE21F1716F1
[Aug 31 08:31] hash3 = F9034C494D5A28C8
[Aug 31 08:31] hash4 = 75DBBC01ED030E0E
[Aug 31 08:32] hash5 = F585034046BE90F7
[Aug 31 08:33] hash6 = 3EBCB6D5E1805F1C
[Aug 31 08:34] hash7 = 7E571EC7FA97D885
[Aug 31 08:34] MD5 error reading PRP proof interim residues file.
[Aug 31 08:34] Waiting 5 minutes to try proof generation again.
[Aug 31 08:34] Waiting five minutes before restarting.

jwnutter · 2020-08-31, 13:43

Quote:

Originally Posted by Viliam Furik

One of them is prime.log, the other is prime.txt. I think the one with a blue square in the icon is the .log

I checked again, and you are correct. Thanks!

2020-08-30, 19:49	#178
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 31×239 Posts	Build 4 Build 4 is available. Go to the first post in this thread to download. This version features an overhaul of the CERT assignment strategy. Looking at the (Ben Delo) data it seems the default settings did not assign CERTs in roughly the same proportion as results. produced. Ben was doing about 2 CERTs for every 3 results. This was due to prime95 requesting making the same number of get cert assignment calls each day regardless of CPU power. That is, a slow dual core CPU would get, on average, the same number of CERT assignments each day as a 24-core AWS server. Build 3 would check for an assignment at startup, every 6 hours, after each result reported, and when sending new expected completion dates. Build 4 will check for an assignment at startup and every N hours where N depends on the number of CPU cores: Code: 20+, N=3 12-19, N=4 7-11, N=6 3-6, N=8 2 N=12 1 N=24 This should lead to a more equitable distribution. Other changes: 1) Getting PRP-CF CERTs is disabled by default. To turn it on, put CertMinExponent=0 in local.txt. The default is 50 million. I've set my computers to accept PRP-CF CERT assignments which should easily handle the load. 2) If you are assigned a PRP-CF CERT, prime95 will get up to 5 at a time. They are quick and this might minimize interruptions. 3) If you set CERT CPU percentage to 50 or more in the advanced resources dialog, then you will be considered a "CERT volunteer". Prime95 will check for CERT assignments every half hour even if you already have a CERT assignment, and will get multiple CERT assignments at a time. 4) You will not get CERT assignments on exponents over 200 million unless you have configured prime95 with an average of 4 or more threads per worker. There are local.txt options to override the get assignment frequency and the quantity that can be queued up at one time. If there is a good reason to make these options public, I'll add them to undoc.txt.

2020-08-29, 09:15	#177
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 724₈ Posts	On a lot of machines and fore some versions now (I cannot recall if it ever was different), Prime95 never was able to restore the maximized status of it's (only) subwindow.

2020-08-31, 10:56	#180
keisentraut Jul 2020 2²×3 Posts	Another issue: Here is an example of Ben Delo having certified his own result: M99841073. I do not understand the math enough, but would this enable Ben Delo to "cheat"? Maybe the server should not assign CERT jobs to the user did the PRP (or "Anonymous"). (PS: Of course I trust Ben Delo not to "cheat". But someone else might.)