Number of PRP tests being reserved

[MikeH]

I have noticed that the rate at which new PRP tests are being reserved has increased significantly, but the effect is no where near as significant in the project stats.

For the last couple of months the leading edge of PRP tests, Max n on this page (http://www.seventeenorbust.com/stats/rangeStatsEx.mhtml), has been moving forward at about 5000 per day. On the 9th of March this suddenly stepped up to ~8000/day.

The project stats graph (http://www.seventeenorbust.com/stats/) shows an increase but it's ~10% not the 60% we're looking for.

I guess there's any one of a number of contributing factors, here's what I can think of.

1. Someone is testing out a caching proxy or client, thus the increased rate is not real :(
2. We have just picked up a lot of new PCs which are either slow or have intermediate block reporting turned off.:)
3. A prime has been found, and the stats page is just moving the numbers for that k forward but not issuing tests (trying to hide the facts?) :D . Only problem with this theory is that even the heaviest k (55459) would only cause ~20% increase. Again, not enough. Maybe it's two primes ?:rotfl:

Maybe it's a combination of the above.? Or maybe someone can enlighten me :confused:

Whatever's happening this page (http://www.aooq73.dsl.pipex.com/ui/9999.htm#S1.2) show that all the sieving we've done over 40T is currently saving ~10 PRP tests per day, and factoring adds a few more to that :cheers:

[Nuri]

I wish it was due to a (couple of) prime(s), but .....


PS: Mike, I was thinking you stopped supporting all users stats page. But, it seems that only the link has changed. I'm happy to find it again. It's very useful for the big picture. :thumbs:

[MikeH]

PS: Mike, I was thinking you stopped supporting all users stats page. But, it seems that only the link has changed. The 'all users' page can be found from either of the 'total' links on this page (http://www.aooq73.dsl.pipex.com/scores.htm). I had to shuffle the number up because real users ids were aproaching 7999!

EDIT: Now with a page ref that's not on my local drive:blush:

[Death]

Originally posted by MikeH
The 'all users' page can be found from either of the 'total' links on URL=C:\tmp\sob\SieveScore\scores.htm this page/URL. I had to shuffle the number up because real users ids were aproaching 7999!

Are you sure? =))

[MikeH]

3. A prime has been found, and the stats page is just moving the numbers for that k forward but not issuing tests (trying to hide the facts?) . Only problem with this theory is that even the heaviest k (55459) would only cause ~20% increase. Again, not enough. Maybe it's two primes ? The lack of comment from either Louie, Dave or Mike is interesting:thumbs:

[jjjjL]

Originally posted by MikeH
The lack of comment from either Louie, Dave or Mike is interesting:thumbs:

< me checks for prime >

Nope. I'm not pulling any strings behind the scenes on this one.

DC users are great. Only they get *excited* when their site is somehow broken or malfunctioning strangely. :)

I have no idea what is causing this. My Guess(tm) is that it may have something to do with the extended expiration time for tests. I kicked it up to 30 days back before Christmas. Perhaps more tests are finishing that would otherwise have been reassigned. This wouldn't produce more stats, but it would produce more finished tests. However, if you are correct with your % jump then this effect could not explain it alone. I did a calculation once and determined that less than 10% of work is "wasted" on tests that never finish.

Maybe linux users are coming back now that v1.2.5 is stable. There's no reason they couldn't roll out some serious production now that the client is solid.

Another way production could be hidden would be if someone were finishing a bunch of tests that were paused. The tests are simply plotted on a per test basis, not per block, so if people are taking breaks in the tests but still finishing them, the stats will look artificially low because current production will be back-averaged over the length the test has been out. In that sense, the stats are smoothed out over time. So an increase in production like this would just show the entire production graph raising, not just the leading edge. Is this happening?

Cheers,
Louie

[dmbrubac]

All I know for sure is factoring is getting swamped! I finished a range while tests were being handed out just 2000 less than the end of the range. I'd previously calculated that I'd have nearly a week before tests entered my range.

What the heck is going on?

[Mystwalker]

(Mike, could you plot a graph showing the daily lower bound of the primary search window - i.e. Max N of the Current Test Window? That would be useful for factorers as well)

Hm, currently, more tests (considering only primary PRPing - no secret or supersecret) get completed than some months ago. Overall production increases again (e.g. look at hadesau (http://seventeenorbust.com/stats/users/user.mhtml?userID=631), who tripled his effort to 6 Mcem/sec) and the amount of new members is above average to my knowledge.
So I think the new client versions (esp. the Linux version) attract new users, plus there are maybe some ppl from other PRPing project (riesel, PSP, 321search, ...) sharing their computing power.

Apart from that, we now have 905 active users and 4068 pending tests. It might be interesting to know how the ratio was in the last months.

[dmbrubac]

I recall that the number of active users was in the 850 range whenever I happened to look at it over the last year or so, but of course, that doesn't speak to the volume of work that each user is doing. I for one am doing only 10% of what I used to, partly to concentrate on factoring.

[MikeH]

I'll do that plot a bit later (not at the right computer right now).

There are new users coming on board, but the bit that confuses me is that there appears to be a serious impalance in the number of tests being reserved vs. the number coming back. A couple of weeks ago, the Max N was moving at about 5000 per day, which equates to about 145 tests per day. The number of tests coming back is about the same (project stats - secret - supersecret), which has been about 200 - 5 - 50. So similar numbers. Today I think we have again moved about 8000, which is about 230 tests. And coming back we have 191 - 4 - 45 = 142.

For this to be 'normal' we would need to have 90 new PCs per day being set up with the client evenly throughout the last 8 days. Most of them with intermediate block reporting turned off, and all of them slower than P4 1.7G (which takes 6 days to complete a typical new test).

Something just isn't right here. :confused:

Louie, is there any way to see if one user or IP address has been reserving lots of tests. Maybe we have a rogue client out there which is crashing and requesting a new test every 15 minutes.

Still, the prime theory was good.:jester:

[dmbrubac]

next.txt was at 561000 this AM and is now at 568000, which is a whole lot better, but I concur, Louie, can you do a query of tests vs production to see if someone is sucking them up?

[Alien88]

I know what it was... one of my newer linux test clients, which worked fine for a while, has issues with getting new blocks and keeps retrying every 1200 seconds. I forgot I still had the client running on a local box, so it was retrieivng a new block every 20 minutes for the past week or so. So that'd be 72 blocks a day it was retrieving. I expired all of them now and killed the client... whoops.

--
Mike

[dmbrubac]

Thanks Alien88. Louie, it probably wouldn't be a bad idea to include the number test for a user in the data Mike gets. It should be pretty easy to find runaways that way.

[Joe O]

In spite of the "runaway client" the PRP boundary has only advanced an average 5400/day since 2003-10-31. This average was at a low of 5300/day 3 weeks ago, and was as high as 5500/day in January.

[jjjjL]

Originally posted by dmbrubac
Thanks Alien88. Louie, it probably wouldn't be a bad idea to include the number test for a user in the data Mike gets. It should be pretty easy to find runaways that way.

This number isn't in the factors file but it is on the http://www.seventeenorbust.com/stats/textStats.mhtml page. So in theory, anyone could have noticed that Mike had way too many tests.

An explaination of what this dump means is well hidden on our site:
http://www.seventeenorbust.com/help/textStats.mhtml

A few users have made alternate stats systems using this data, such as http://sob.qik.nl/ and http://marcc.no-ip.org/SoB/Teamstats.php . A few others use it as well.

Cheers,
Louie

[MikeH]

Maybe we have a rogue client out there which is crashing and requesting a new test every 15 minutes. What a guess (even if I do say so myself):D

So the Max N now won't move for the next 2-3 days.:spank:

[MikeH]

No so relevent now, but here's that chart .....

[Mystwalker]

Thanks for the chart, Mike! :cheers:
So basically, one can assume ~5,000 increase per day. I calculated with 10,000/day before, as I didn't know the correct value...

[Joe O]

Originally posted by Joe O
In spite of the "runaway client" the PRP boundary has only advanced an average 5400/day since 2003-10-31. This average was at a low of 5300/day 3 weeks ago, and was as high as 5500/day in January.
The average advance of the PRP boundary since 2003-10-31 has dropped back to 5300/day.
The average advance for the last week was 3700/day.

[Keroberts1]

I'm not sure of the numbers but isn't it a bit more than that lately? Do we have another runaway client or are we just finishing more tests. The total equivalent processor power has bee nincreasing steadyily for a while now. Approaching the levels we had before the site went down for a few days a few months ago.

[ceselb]

I think ars technica is ramping up a bit to catch up to anandtech.

[Keroberts1]

thats great and i wish a prime upon them in return

[MikeH]

Proth tests completed today: 217
- secret tests: 3
- supersecret tests: 14

That's 200 new tests completed in the last 24 hours. :|party|:

Many thnaks to all those that have brought new resources along to SoB. :thumbs:

[Keroberts1]

so how fast does that mean the prp line is moving?

[MikeH]

so how fast does that mean the prp line is moving? For current sieved levels, one test pushes PRP forward by 1M/29000 = 34.5

So for 200 per day, thats ~6900, which is a lot better than we've been doing of late.

[Joe O]

The average advance for the last week was 5838/day.

[Keroberts1]

looks like the prp tests are getting done at a higher rate again.

291 today -71 supersecrets - 7 secrets = 213 new prp tests.

Very positive looking.

A prime or 2 adn we'll be moving at 10K a day again.