George, I don't know if this is expected behavior or not, but to report...

Overnight my Winblows box did two Certs. [URL="https://www.mersenne.org/report_exponent/?exp_lo=97122001&full=1"]97122001[/URL] and [URL="https://www.mersenne.org/report_exponent/?exp_lo=97513771&full=1"]97513771[/URL]. Both reported back to the server, and received a "Certificate did not match" message. I see that you then did a Cert run on them.

Two observations...

1. It would be cool if the server could report (on the Web UI) if the cert run was successful, or not (I assume that's underway; or at least modeled).

2. This machine has been running DCs for years without issue.

Thoughts?

Edit: Just completed a third CERT run. Again, the server reported a non-match. George: Let me know if you want the Prime95 log file etc.

[QUOTE=chalsall;551555]Edit: Just completed a third CERT run. Again, the server reported a non-match. George: Let me know if you want the Prime95 log file etc.[/QUOTE]

Too late to edit my above...

I just remembered that the comms for this machine is going through the GPU72 Proxy. I don't /think/ this should be causing a problem, but to put it on the table.

Edit: I just removed the Proxy setting in prime.ini, to eliminate this variable. Hopefully, this machine will be assigned another CERT job to test this.

[QUOTE=chalsall;551555]Overnight my Winblows box did two Certs. [URL="https://www.mersenne.org/report_exponent/?exp_lo=97122001&full=1"]97122001[/URL] and [URL="https://www.mersenne.org/report_exponent/?exp_lo=97513771&full=1"]97513771[/URL]. Both reported back to the server, and received a "Certificate did not match" message. I see that you then did a Cert run on them.[/QUOTE]Chris: Hmm, I'm now 6/6 Cert successes on Windows (3 different systems; all Windows 10). Clearly you should ASAP try a comparative run on linsux, same proxy arrangement, and without proxy.

George,
could there be something meaningful entered in the result field for Cert runs on those exponent status pages? And on user results summary page?
If nothing else, "match" vs. "not a match" or something similar. Without it, we'd need to check each Cert result's exponent page manually to see what challsall and I are reporting.
An mprime build for test would be good too.

[QUOTE=chalsall;551555]]
Overnight my Winblows box did two Certs. [URL="https://www.mersenne.org/report_exponent/?exp_lo=97122001&full=1"]97122001[/URL] and [URL="https://www.mersenne.org/report_exponent/?exp_lo=97513771&full=1"]97513771[/URL]. Both reported back to the server, and received a "Certificate did not match" message. I see that you then did a Cert run on them.

Two observations...

1. It would be cool if the server could report (on the Web UI) if the cert run was successful, or not (I assume that's underway; or at least modeled).

2. This machine has been running DCs for years without issue.

Thoughts?[/QUOTE]


My certifications did match the proof.

I think your machine either has a hardware issue, which you believe is unlikely, or there was trouble downloading the initial residue. I think the version you are running does not do an MD5 hash comparison on the starting value. The unreleased current version does.

The reported results via proxy look fine, just didn't match. There has been no testing of getting the initial value through a proxy server. You might try turning on Debug=2 in the [Primenet] section of prime.txt. It might crash or it might provide interesting info.

BTW, all re-certifications are assigned to me. To prevent a bad actor from taking delight in marking good proofs bad, one cannot return a bogus cert result and hope to get the assignment again (even under a different user id) and return a matching bogus cert result.

[QUOTE=Prime95;551580]You might try turning on Debug=2 in the [Primenet] section of prime.txt. It might crash or it might provide interesting info.[/QUOTE]

OK... I stopped (and exited) Prime95, set debug to 2, and relaunched. Luckily it was immediately given a Cert assignment, so we'll see what happens in the next couple of hours.

I've left the Proxy off for the time being. Change one variable at a time during debugging.

Edit: Oh, and you've got some debugging to do with the ETA calculations on the server. The client is saying 45 minutes, while the Assignments page is saying 5 days... :smile:

[QUOTE=chalsall;551582]Luckily it was immediately given a Cert assignment, so we'll see what happens in the next couple of hours.[/QUOTE]

OK... The Cert job just finished, and this time it reported a successful match.

Thus, very likely an issue with the Proxy. I'll drill down on the logs tomorrow to see what I can see. Heading off to the Hash shortly...

If a PRP with proof is a DC and the res64 matches the first PRP run, is the verification of the PRP DC proof performed?

How does one submit a proof file from a completely manual PRP run?
(No primenet.py use, no python on the system)
Dropbox and email?

[QUOTE=kriesel;551593]If a PRP with proof is a DC and the res64 matches the first PRP run, is the verification of the PRP DC proof performed?

How does one submit a proof file from a completely manual PRP run?
(No primenet.py use, no python on the system)
Dropbox and email?[/QUOTE]

If a PRP-DC has a proof, I think it should be verified as an extra guarantee. I personally feel a certified proof is a stronger DC than 2 matching 64-bit residues.

I have a standalone uploading program I can make available.

[QUOTE=Prime95;551596]I have a standalone uploading program I can make available.[/QUOTE]

Gpuowl's python uploader should also work.

[QUOTE=Prime95;551596]If a PRP-DC has a proof, I think it should be verified as an extra guarantee. I personally feel a certified proof is a stronger DC than 2 matching 64-bit residues.[/QUOTE]I think the verify run is worthwhile in such a case too. And that the proof math is sound, and more resistant to user error or misbehavior than the status quo.
[QUOTE]I have a standalone uploading program I can make available.[/QUOTE]That would be good. As things stand, I'll be accumulating proof files on a local 2TB shared drive. Don't know where / how to upload them, but definitely know they're valuable. (Hmm, I better initiate backups for that drive soon.)

When might a proof-enabled build of mprime become available? Some of us are running PRP cpu sessions on Colab. Those are almost a waste since they lack proof code, necessitating a full PRP DC.

When might a prime95 V30.1b2 come out? (I just finished rolling v30.1b1 out to all running systems here. Hoping I don't get inundated by Cert work displacing my own goals.)

Try [url]https://www.dropbox.com/s/4h7jujxwzmhcsmt/uploader.zip?dl=0[/url] for Windows

Launch from a DOS shell. Command line arguments are user_id, proof_file_name, chunk_size, bandwidth_limit

Default chunk size is 5 (MB)
Default bandwidth_limit is unlimited. I'm using 1 (Mbps)