Is pfactor work type broken?

[Prime95]

I just changed the server to return pfactor assignments with 1.3 PRP tests saved rather than 2.0 PRP tests saved.

e.g:

PFactor=9A0F34B929B282C41C75682C9496D92B,1,2,109343671,-1,76,1.3

Let me know if this breaks any versions of prime95, gpuowl, or mlucas.

[Zhangrc]

Great!
Now I can complete 5 P-1 assignments every day now!
PS: you haven't marked the AID

[Prime95]

Originally Posted by Zhangrc

Great!
Now I can complete 5 P-1 assignments every day now!
PS: you haven't marked the AID

I unreserved the assignment, AID is now useless.

[techn1ciaN]

Originally Posted by Prime95

I just changed the server to return pfactor assignments with 1.3 PRP tests saved rather than 2.0 PRP tests saved.

Any reason for that particular number?

[Zhangrc]

Originally Posted by Prime95

I just changed the server to return pfactor assignments with 1.3 PRP tests saved rather than 2.0 PRP tests saved.

e.g:

PFactor=9A0F34B929B282C41C75682C9496D92B,1,2,109343671,-1,76,1.3

Let me know if this breaks any versions of prime95, gpuowl, or mlucas.

You haven't changed the number on https://www.mersenne.org/workload/ .

The number 1.3 is a rough estimate, because most of the exponents are verified by a PRP and a CERT, which need 1.01 tests, but some need 2 PRP tests (without proof) or 2.02 LL tests. That's about 1.05 tests on average, but 1.3 is still reasonable.

However, I think the best idea is to let PRP users run 1-test-saved bounds, while P-1 users should run larger bounds.

[LOBES]

Originally Posted by Prime95

I just changed the server to return pfactor assignments with 1.3 PRP tests saved rather than 2.0 PRP tests saved.

e.g:

PFactor=9A0F34B929B282C41C75682C9496D92B,1,2,109343671,-1,76,1.3

Let me know if this breaks any versions of prime95, gpuowl, or mlucas.

Thanks for this change. I was manually loading P-1 assignments and changing tests_saved from 2 to 1. I can now just let prime95 run and grab its own since 1.3 is low enough for me.

I was thinking about requesting a new prime.txt parameter to allow you to set the P-1 tests_saved value. If absent (or incorrectly set) it would just default to the server value. Otherwise would use the value set. I'm not sure how complicated it would get when multiple workers, etc were in play and users wanted different values for each one.

All of my machines are very simple set-ups so I tend to overlook the difficulties it would be for others.

[DrobinsonPE]

My automatically assignments still show up as 2 instead of 1.3

This one was assigned on the 12th which is still today in PST timezone.
PFactor=obscured,1,2,110557939,-1,77,2

[tdulcet]

Originally Posted by Prime95

Let me know if this breaks any versions of prime95, gpuowl, or mlucas.

This unfortunately broke our PrimeNet script, as it uses a regular expression to parse the assignment lines, which was expecting an integer. However, I just pushed a simple update to fix it. See here for the full changes. Anyone who is using our PrimeNet script with P-1 assignments will need to update as soon as possible.

This may also break Mlucas and GpuOwl v6.11, which from looking at their source code are both expecting unsigned integers.

Originally Posted by LOBES

I was thinking about requesting a new prime.txt parameter to allow you to set the P-1 tests_saved value. If absent (or incorrectly set) it would just default to the server value. Otherwise would use the value set.

It would be relatively trivial to add this to our PrimeNet script (for Mlucas and GpuOwl users) if there was sufficient interest.

[Prime95]

Originally Posted by DrobinsonPE

My automatically assignments still show up as 2 instead of 1.3
PFactor=obscured,1,2,110557939,-1,77,2

I'm not seeing that here. I got an assignment in 30.8b10 and it appears as xxxx,1.3 in worktodo.txt.

[DrobinsonPE]

Originally Posted by Prime95

I'm not seeing that here. I got an assignment in 30.8b10 and it appears as xxxx,1.3 in worktodo.txt.

Ok, I understand.

The worktodo.txt on that computer shows the Pfactor assignments ending in 1.3. However on my Mersenne.org/workload web page the same assignments for the same computer show up ending in 2 not 1.3.

I was not looking at the worktodo file on my computer when I made the previous comment.

[ixfd64]

Originally Posted by Zhangrc

However, I think the best idea is to let PRP users run 1-test-saved bounds, while P-1 users should run larger bounds.

Seconding this. I have some less-active computers that are running exclusively P-1 assignments. These computers can afford to use larger bounds.