Proposed DC assignment and recycle rules

[cuBerBruce]

I have a machine that has been doing a DC of M32062099. It was supposed to have 100 days to complete the assignment, and it is over 80% done and well on track of completing the assignment within the 100 days. However, it appears that for some reason within the last couple days, Primenet seems to have considered it to have expired and assigned it to someone else. I would like to know why this happened. I could have moved this assignment to a faster machine if I had known the 100-day time frame would not be honored.

[chalsall]

Originally Posted by cuBerBruce

I could have moved this assignment to a faster machine if I had known the 100-day time frame would not be honored.

Hmmm... Strange. And you definitely were reporting progress regularly:

32062099 D            2   68 2014-06-12 2014-04-05 2014-04-04 2014-04-03 cuBerBruce CenTurion
32062099 D LL, 7.90%  9   61 2014-06-12 2014-04-12 2014-04-11 2014-04-03 cuBerBruce CenTurion
32062099 D LL, 13.50% 13  50 2014-06-05 2014-04-16 2014-04-15 2014-04-03 cuBerBruce CenTurion
32062099 D LL, 19.30% 17  40 2014-05-30 2014-04-20 2014-04-19 2014-04-03 cuBerBruce CenTurion
32062099 D LL, 31.50% 24  36 2014-06-02 2014-04-28 2014-04-27 2014-04-03 cuBerBruce CenTurion
32062099 D LL, 32.90% 27  37 2014-06-06 2014-05-01 2014-04-30 2014-04-03 cuBerBruce CenTurion
32062099 D LL, 34.70% 29  37 2014-06-08 2014-05-03 2014-05-02 2014-04-03 cuBerBruce CenTurion
32062099 D LL, 40.40% 33  36 2014-06-11 2014-05-07 2014-05-06 2014-04-03 cuBerBruce CenTurion
32062099 D LL, 47.50% 38  34 2014-06-14 2014-05-12 2014-05-11 2014-04-03 cuBerBruce CenTurion
32062099 D LL, 55.90% 44  32 2014-06-18 2014-05-18 2014-05-17 2014-04-03 cuBerBruce CenTurion
32062099 D LL, 67.00% 52  24 2014-06-18 2014-05-26 2014-05-25 2014-04-03 cuBerBruce CenTurion
32062099 D            2   10 2014-06-14 2014-06-05 2014-06-04 2014-06-02 Lazer13    Pappa1

[chalsall]

Originally Posted by chalsall

Hmmm... Strange.

Actually, I just realized... Not strange. For "Category 1" DC assignments you're only given 60 days, not 100.

You should change your "Days to Queue" value to be larger than 10 to get Cat 2.

[cuBerBruce]

Originally Posted by chalsall

Actually, I just realized... Not strange. For "Category 1" DC assignments you're only given 60 days, not 100.

You should change your "Days to Queue" value to be larger than 10 to get Cat 2.

The machine isn't even fast enough to receive a Cat 1 assignment. And if it was Cat 2 when assigned, it should have had at least 100 days to work on it even if the exponent became within the Cat 1 range:

Assignments are recycled when the exponent moves into the first category and the assignment is more than 100 days old.

[chalsall]

Originally Posted by cuBerBruce

And if it was Cat 2 when assigned, it should have had at least 100 days to work on it even if the exponent became within the Cat 1 range:

Agreed -- iff it was indeed a Cat 2 when assigned. I don't have the data to answer that, but if it was then you are correct, and there's a bug in the Primenet recycling code.

[TheMawn]

Does Prime95 delete the checkpoints if the assignment is recycled? It would be a damn shame to lose all that work. If the checkpoints are still there, I guess cuBerBruce could just finish off the job.

[chalsall]

Originally Posted by TheMawn

Does Prime95 delete the checkpoints if the assignment is recycled? It would be a damn shame to lose all that work. If the checkpoints are still there, I guess cuBerBruce could just finish off the job.

We've got a race-condition on this though -- it looks like Lazer13's Pappa1 machine is pretty fast (or, at least, claims to be based on the estimated completion time).

But, if cuBerBruce wants to (and I don't think it would be inappropriate given the circumstances), I don't believe the checkpoint files are deleted -- simply the entry in the worktodo.txt file. Thus, simply adding "DoubleCheck=N/A,32062099,70,1" to the top of the file and restarting prime95/mprime, the test should continue from where it last stopped.

If cuBerBruce beats Lazer13, then the next time Lazer13 machine checks in the client will be told to stop the work, and be given another assignment to work on. If Lazer13 beats cuBerBruce, then cuBerBruce will still complete the assignment (most likely uselessly, unless it turns out a TC is going to be needed).

Whatever happens, this is an unfortunate situation, and will most likely result in someone wasting cycles. And if it is in fact a bug with the recycling code it should be fixed ASAP.

[cuBerBruce]

t just seems strange to me that if it was Cat 1 when assigned, I would not have noticed that and realized I would probably have to eventually transfer it to another computer in order complete it. Unless I forgot the limit was 60 days and not 90 days for Cat 1 DC, I might have thought I was fine.

It looks to me that the Cat 1/Cat 2 boundary might have been somewhere in that area at the time. Perhaps someone knows what the Cat 1/Cat 2 boundary was on April 3?

For what it's worth, I do have checkpoint files. (In fact, I haven't stopped it from running yet, though Prime95 no longer considers it a valid assignment.)

I just saw chalsall's suggestion on restarting it. If I do restart it in this manner, I am wondering if I should do it on my Sandy Bridge machine rather than the computer it's been running on.

[owftheevil]

The cat1 assignments I got in early April were < 32000000, circumstantial evidence that this was a cat 2 assignment.

[chalsall]

Originally Posted by cuBerBruce

t just seems strange to me that if it was Cat 1 when assigned, I would not have noticed that and realized I would probably have to eventually transfer it to another computer in order complete it. Unless I forgot the limit was 60 days and not 90 days for Cat 1 DC, I might have thought I was fine.

Possible.

Originally Posted by cuBerBruce

It looks to me that the Cat 1/Cat 2 boundary might have been somewhere in that area at the time. Perhaps someone knows what the Cat 1/Cat 2 boundary was on April 3?

I don't know if that data is being recorded by Primenet and/or others. I'm not spidering that data, so I can't speak to the question.

Originally Posted by cuBerBruce

For what it's worth, I do have checkpoint files. (In fact, I haven't stopped it from running yet, though Prime95 no longer considers it a valid assignment.)

Now, that's also strange in my mind. What version of Prime95 are you running? I thought that once told "no valid AID" Prime95 would stop work. I'm pretty sure this is the case for recent versions of mprime (but I could be wrong).

Originally Posted by cuBerBruce

I just saw chalsall's suggestion on restarting it. If I do restart it in this manner, I am wondering if I should do it on my Sandy Bridge machine rather than the computer it's been running on.

The faster the better. Also, George might want to consider contacting Lazer13 via email, explain the situation, and suggest they not start the work.

Overall, I'd say this has been a worthwhile exercise, even if a few GHz days might be wasted by someone...

There might be a bug on Primenet which should be fixed if real. Or, there might have been an "error between keyboard and chair" (read: people who have committed to the new rules should be fully aware of the "contract" they enter by doing so).

Either way, valuable in my mind.

P.S. We all really need to get out more....

[cuBerBruce]

OK, I've gone ahead and moved the assignment to my Sandy Bridge machine, and continued running it there. The machine also had a DC that was 97% done that was getting close to crossing into Cat 1 territory (74 days old). I transferred that to the Sandy Bridge as well, for good measure, using the proper AID.

Originally Posted by chalsall

Now, that's also strange in my mind. What version of Prime95 are you running? I thought that once told "no valid AID" Prime95 would stop work. I'm pretty sure this is the case for recent versions of mprime (but I could be wrong).

Oops, I thought the communication thread had stopped submitting expected completion date info on that exponent, but that is not true. It still communicated an expected completion date for it late this morning. So Prime95 seemed to be acting like it thought it was still a valid assignment.

The Prime95 version is V27.9, build 1.