P-1 coordination thread discussion

[dmbrubac]

I know things are going fast but that seemed REALLY fast. Could there be another runaway?

[Mystwalker]

Well, production went to by more than 50% in the last 2 months. Even considering cEM/s is increasing for higher n's and those in turn take longer, there should be a massive performance plus in the PRP section...

[Nuri]

Originally posted by dmbrubac
I know things are going fast but that seemed REALLY fast. Could there be another runaway?

I have no idea on the level of difficulty in making such a query, but a quick analysis (by someone who has access to the server/tests data) of number of pending tests vs. number of daily completed tests by each user would give a clear picture in case there is a problem.

[Keroberts1]

the number of pending tests has gone through the roof i think it would be a good idea to check if there is another run away soon it certinly does appear that way.

[dmbrubac]

next.txt moved by more than 12000 over the last 24 hours! There is definitely a run-away here somewhere. Runaway detection and identification should be fairly easy for the server, no?

[Nuri]

We've finished 275 first time PRP tests within the last 24 hours.

This is roughly 9500 per day (=1,000,000/29,000*275).


There is an unexplained figure of 72 tests per day (=(12000-9500)*29,000/1,000,000).

72 tests per day =1 test per 20 minutes.

This definitely is the same problem we've encountered before.

A single PC of one of the users is continuously grabbing an additional PRP test every 20 minutes, without doing any work on them.

[Nuri]

Using 47 instead of 48 will bring you additional factors in the p range of 141T to 281T. Since only 15% of the factors are left in that range (see http://www.aooq73.dsl.pipex.com/scores_p.htm), I doubt if it would be worth using 47.

If you have 256 MB ram, I would recommend you not to use anything above 200, or your system will slow down or crash, especially if you are using the PC for some other purposes as well.

[jjjjL]

Originally posted by Nuri
We've finished 275 first time PRP tests within the last 24 hours.

This is roughly 9500 per day (=1,000,000/29,000*275).


There is an unexplained figure of 72 tests per day (=(12000-9500)*29,000/1,000,000).

72 tests per day =1 test per 20 minutes.

This definitely is the same problem we've encountered before.

A single PC of one of the users is continuously grabbing an additional PRP test every 20 minutes, without doing any work on them.

I checked to see how many tests were out to each user id:

userid tests
| 4056 | 30 |
| 6994 | 33 |
| 5446 | 33 |
| 4494 | 33 |
| 6494 | 33 |
| 2933 | 38 |
| 4116 | 42 |
| 6555 | 43 |
| 1800 | 43 |
| 6749 | 46 |
| 5965 | 50 |
| 6564 | 58 |
| 2634 | 69 |
| 4396 | 82 |
| 6251 | 85 |
| 1269 | 91 |
| 50 | 119 |
| 6696 | 128 |
| 2891 | 184 |
| 5218 | 202 |
+--------+-------+

hmmm, most those make sense except 5458. He has only finished 6 tests. But the strange thing is, he has reported progress on all the tests he has out. I dunno what he's up to. Mike will have to get in touch with him to make sure its all on the up and up.

Cheers,
Louie

[Keroberts1]

5458?

I assume yo umean 5218 adn his rate is extremely low. I expect that hismachine is simply malfunctioning or he's have a much higher equivalent power and cem/s rate.

[Keroberts1]

how much memory is everyone using to run these tests? just wondering because I've noticed in my own experiments taht if yo udon't have a large ammount of memory devoted to the second portion of the test then the likely hood of finding a prime is severly impared. The same series of testswith 64 mb of ram gives like a .070 primes in a range and 256 gives .159. Are some people using 64 Mb of ram?

[hc_grove]

Originally posted by Keroberts1
how much memory is everyone using to run these tests? just wondering because I've noticed in my own experiments taht if yo udon't have a large ammount of memory devoted to the second portion of the test then the likely hood of finding a prime is severly impared. The same series of testswith 64 mb of ram gives like a .070 primes in a range and 256 gives .159. Are some people using 64 Mb of ram?

From the very beginning Louie has recommended using at least 128 MB of RAM, so I don't hope anyone is only using 64.

Personally I use 384 on one machine and 512 on another.

[Troodon]

Keroberts1 , please note that sbfactor sets the optimal bounds taking in account the memory available. AFAIK, if you give it more memory, it will increase the bounds, and you will get more chances of finding a factor, but at the same time it will take more time to perform a test. If you manually mantain the bounds but give it more memory, the tests will be faster (only if the previosuly given amount of memory was less than the maximum needed).

[dmbrubac]

<rant>
Is anything being done about this? That machine has more than 200 test and even if there is a minor trickle of work coming out, I believe it constitutes abuse. The tests should be forcibly expired and the IP banned. </rant>

I continued factoring ranges that had been passed in the belief that a large portion of the tests would be returned to the active heap - and I found a few factors that didn't score well. If it trns out an expired test has had a factor found for it, will the test be removed and the factor credited?

The PRP wave is still moving a) faster than it should for the tests being returned and b) faster than we can handle.

There are two things we need desperately:
1) a run-away warning system. A query that compares number of tests versus work done would generate candidates that could then be manually inspected.
2) more factorers. Real speed has increased we are being swamped. Could a call be put out to the sievers? There is already a discussion in progress re stopping sieving altogether.

[Frodo42]

A run-away warning system would be a good thing.

I don't thinkt there is all that much help to get from sievers, I think the machines used for sieving mostly is to slow/old to use for factoring, but i may be wrong and it would be nice to be able to keep up with prp'ing.

[Alien88]

Well, the guy with the runaway tests never responded to my emails, so I expired his tests.. the client seems fine now, so who knows.

I also when through a few other accounts who had a lot of pending tests and expired ones that were 0% no progress in 3 weeks.

[dmbrubac]

I don't normally post this kind of stuff, but;

Woo Hoo :thumbs:

[MikeH]

I also when through a few other accounts who had a lot of pending tests and expired ones that were 0% no progress in 3 weeks. Great job Mike. Maybe you could put together a little script and run it periodically. :cheers: