Getting CERT assignments

[James Heinrich]

Originally Posted by harlee

James, you're probably already doing this, but have you looked CERT results being returned to see which version is the highest that are returning CERT work?

I have just now, I examined all the results for 2022-Dec-01 through 2022-Dec-31, counting both the total number of LL/PRP results submitted, and the number of CERT results, by software version:

December 2022 (entire month)
Prime95/mprime version  LL/PRP  CERT
v23,(aka WZ1)           =    1  
v24 final,(aka Wc1)     =    3
v25.11,build 2          =   22
v25.9,build 1           =    1
v25.9,build 4           =   12
v26.5,build 5           =    4
v26.6,build 3           =   74
v27.7,build 2           =   23
v27.9,build 1           =   78
v28.10,build 1          =  117
v28.5,build 2           =   25
v28.6,build 1           =   71
v28.7,build 1           =   90
v28.9,build 2           =  123
v29.1,build 13          =    8
v29.1,build 14          =    1
v29.1,build 16          =   16
v29.2,build 5           =    8
v29.3,build 1           =   16
v29.4,build 4           =    1
v29.4,build 5           =   46
v29.4,build 7           =  125
v29.4,build 8           =  644
v29.5,(aka Wh4)         =    1
v29.5,(aka Wh8)         =    1
v29.5,build 6           =    1
v29.5,build 9           =    1
v29.6,build 3           =    1
v29.6,build 7           =    6
v29.7,build 1           =    3
v29.8,build 1           =    2
v29.8,build 3           =  250
v29.8,build 5           =   68
v29.8,build 6           =  599
v29.8,build 7           =  105
                        
v30.10,build 1          =    1
                        
v30.3,build 2           =    3     19
v30.3,build 3           =   13     25
v30.3,build 4           =    1      7
v30.3,build 5           =    2     11
v30.3,build 6           = 2728   8376
                        
v30.4,build 8           =   13
                        
v30.5,build 2           =   52     32
                        
v30.6,build 2           =   12
v30.6,build 4           =  661    353
                        
v30.7,build 7           =  384     90
v30.7,build 8           =   75    180
v30.7,build 9           = 5583   3845
                        
v30.8,build 3           =    1      8
v30.8,build 5           =    5     16
v30.8,build 6           =    8     13
v30.8,build 7           =    1
v30.8,build 10          =   92     41
v30.8,build 11          =   81      3
v30.8,build 12          =   23     27
v30.8,build 14          =   62
v30.8,build 15          = 1120     48
v30.8,build 16          = 1203    152
v30.8,build 17          = 2562    165
v30.8,build 18          = 1640     13
                        
v30.9,build 1           =    7      1
v30.9,build 3           =   91     10

(as an aside, I'm astounded to see some truly ancient versions still turning in results)

But looking at more recent versions, you can see that the most popular version of v30.3 and v30.7 have a vaguely-similar magnitude of PRP and CERT results, but the popular recent versions of v30.8 do not. Based on the above numbers alone, I'd guess the CERT-breaking change happened somewhere around 30.8b13.

[slandrum]

Originally Posted by James Heinrich

(as an aside, I'm astounded to see some truly ancient versions still turning in results)

I've seen assignments that are 7-9 years old being turned in occasionally. Sometimes assignments that start as FTC end up turning in as a quad-check instead.

[Prime95]

According to the prime95 source code there are several reasons the client may never make requests for CERT work.

• CertWork=0 in local.ini
• Using PauseWhileRunning
• Setting WellBehavedWork
• Setting HoursPerDay < 12
• Setting CertDailyCPULimit and previous certs exceeded the quota
• Setting DailyDownloadLimit and exceeding the quota
• Setting QuitGIMPS

I suspect the PauseWhileRunning is what is catching James. I vaguely remember it was James that asked for this :)

[LOBES]

Originally Posted by Prime95

According to the prime95 source code there are several reasons the client may never make requests for CERT work.

• CertWork=0 in local.ini
• Using PauseWhileRunning
• Setting WellBehavedWork
• Setting HoursPerDay < 12
• Setting CertDailyCPULimit and previous certs exceeded the quota
• Setting DailyDownloadLimit and exceeding the quota
• Setting QuitGIMPS

I suspect the PauseWhileRunning is what is catching James. I vaguely remember it was James that asked for this :)

In my case, I changed no configuration (prime.txt or local.txt). I just upgraded prime itself.

If it was configuration, it wouldn't check for certs when I'm simply running prime -m.

[Prime95]

There is a possibility that prime95 has a bug in monitoring daily download limit or cert cpu time. Send me local.txt if you think that may be the case.

[James Heinrich]

Originally Posted by Prime95

According to the prime95 source code there are several reasons the client may never make requests for CERT work.

That's a useful list for reference, but none of it applies. A vanilla installation of mprime v30.8b15 will not request CERT work (except while sitting in the -m menu). A vanilla installation of 30.8b7 will request CERT work occasionally as it should.

Have you looked at the code involving PAUSEABLE_WORKERS_RUNNING that Aexoden mentioned?

edit: here is the code he referenced (commonc.c:6080 in v30.10b1.src)

can_get_cert_work = (header_words[1] & HEADER_FLAG_WORK_QUEUE) && IniGetInt (LOCALINI_FILE, "CertWork", 1) && DAYS_OF_WORK > 0.0;
if (WELL_BEHAVED_WORK || SEQUENTIAL_WORK == 1) can_get_cert_work = FALSE;
if (CPU_HOURS <= 12) can_get_cert_work = FALSE;
if (PAUSEABLE_WORKERS_RUNNING) can_get_cert_work = FALSE;

It's that last line that's suspicious.
I don't speak C very well at all, but to me it looks that in commonb.c, start_pause_while_running_timer() always sets PAUSEABLE_WORKERS_RUNNING=true, and start_pause_while_running_timer() is always called (even if there is no PauseWhileRunning entry configured). I could easily be wrong, but would appreciate your review of the logic.

[Prime95]

Originally Posted by James Heinrich

Have you looked at the code involving PAUSEABLE_WORKERS_RUNNING

Hopefully fixed in 30.10 build 3

[James Heinrich]

I upgraded my worker that I had downgraded to v30.8b7 (and was therefore getting CERT work again) and immediately on starting a worker it seemed to ask for CERT work (didn't get any but the asking is the important bit) so looks promising. Will check back later today or tomorrow to make sure it's completing CERT work as expected.

Thanks George.

[James Heinrich]

Inexplicably, my v30.8b17-Win64 (doing only P-1) just started getting piles of CERT assignments. It saw one CERT on 2022-08-05 [v30.8b15], but before that not since 2022-02-21 [v30.8b10] when it had been getting many but suddenly stopped getting them.

[James Heinrich]

Now I find v30.8b17 perhaps even too enthusiastic in getting CERT assignments...

[Main thread Feb 3 10:40] Stopping all worker windows.
[Main thread Feb 3 10:40] Execution halted.
[Main thread Feb 3 10:40] Choose Test/Continue to restart.
[Comm thread Feb 3 13:18] PrimeNet success code with additional info:
[Comm thread Feb 3 13:18] Server assigned CERT work.
[Comm thread Feb 3 13:18] Got assignment 91642DB2B715CA2DA372617384551685: CERT M5881097
[Comm thread Feb 3 13:18] Done communicating with server.
[Comm thread Feb 3 17:48] PrimeNet success code with additional info:
[Comm thread Feb 3 17:48] Server assigned CERT work.
[Comm thread Feb 3 17:48] Got assignment AE074707EEBEF98795CC7AD0DC963F1E: CERT M5951263
[Comm thread Feb 3 17:48] Done communicating with server.
[Main thread Feb 3 17:53] Starting worker.

I was doing other CPU-sensitive things on the PC at the time so I stopped all workers. While all workers were stopped, Prime95 still contacted the server and managed to secure two CERT assignments (I didn't think multiple concurrent CERT assignments were even allowed?). Once I restarted the worker thread (by chance only 5 mins after the second assignment) it very quickly completed the assignments, but I could've stopped the machine for hours/days. I don't think Prime95 should check for CERT assignments if the target worker is stopped (either manually or via PauseWhileRunning). Getting CERT assignments for a worker that's idle due to lack of available stage-2 RAM would be not only acceptable but desirable.

[Prime95]

You can get up to 5 CERT assignments at one time -- if the exponents are small.

Next release will not get CERT assignments if workers are not active.