Sieve coordination discussion

[vjs]

10% sounds about right, my quick guestimates show my fastest computer down from ~700 kp/s to 650 kp/s....

[vjs]

e

RU a round??

[vjs]

I'm glad to see the renewed sieve effort, A couple of us have been trying to get this done for some time, and I only wish I had personally started sieving earlier.

Had we all started earlier, ie while around 3m, more of the founds factors would have eliminated more tests. Thus far with the sieve effort is almost at 500T and we have eliminated 31% of the factors between 1m and 20m . Even if we find 4 more primes before 20m at least the majority of these factors were/are useful, and the more we find the better. In addition we will more than likely double check more than half of the k's below 20m. Point being even if you find a k/n below 7m you have eliminated a double check. If it's more than the currrent threashold of ~7.35M you will have eliminated 2 checks.

Sad news is there is a point of diminishing returns, good news is we are not there yet. Our current client can sieve upto 1000T so basically we are half way there. If we sieve past 1000T is questionable, In addition there are k/n pairs beyond 20m that we will have to test as well but not for a few years. Hopefully the project will have eliminated alot of k's by then.

My main point is we should try to drive up the sieving T up as fast as possible to some T value and stop sieving.

So how long will it take us to get to 1000T???

Well that of course depends on how many people we have helping...

Here are some numbers. Currently we are finishing around ~1000G per day which is rougly 11,500 kp/s.

At this rate we will get to 1000T in about 16 months.

So how many computers will it take to sieve to 1000T or 1P, in 6 months?

Well if we increased our total sieve speed to the <b>primorial rate</b> of 30029 kp/s we could do it. This could be done with a combination <100 of reasonable xp1800 and P3-1000mhz computers.

If I had a hundred of my fastest sieving computer (a 2500mhz Barton) it would only take about 3 months.

Thank-you all for helping out.

[vjs]

factor=329.299T k=21181 n=4604 score=0.001 Fri 26-Nov 2004 dmbrubac

I guess every factor helps... But wow that's a small n!!!

I think you not only hold the 100K record but that's one of the smallest I've seen with sieve.

Except for Joes P-1 is really tiny

Factor=746.073P k=4847 n=351 score=1.021 Joe_O

[vjs]

Nice range E you too larsivi

Glad to see the regulars reserving larger ranges, I'm sure Mike and cesleb will appreciate it.

[ceselb]

Originally posted by vjs
Glad to see the regulars reserving larger ranges, I'm sure Mike and cesleb will appreciate it.

Yes, I do.

[maddog1]

Speak of the devil...how about this (obtained through sieving)
490586292728209 | 24737*2^1303+1

[larsivi]

Originally posted by vjs
Nice range E you too larsivi

Glad to see the regulars reserving larger ranges, I'm sure Mike and cesleb will appreciate it.

Well, thank you. The point is that except for one slow, noisy machine, I seldomly know what power is available for more than one or two weeks ahead. I therefore try to reserve almost exactly as much as I know I'll be able to process.

[engracio]

Well, thank you. The point is that except for one slow, noisy machine, I seldomly know what power is available for more than one or two weeks ahead. I therefore try to reserve almost exactly as much as I know I'll be able to process.

Ya............



e

[vjs]

Does anyone have any data on how much time processing power each k takes?

I know I did some of this a while ago seeing how much of a speed up we would see from eliminating various k...

Do some k's play well together, are some outstandingly different such as 67907??

[vjs]

I know sometimes my rate is up and down depending upon which machines are active. Yesterday lost a couple slower machines probably 600kp/s total combine that with the temporary high n-sieve effort, I'm probably down to around 2000kp/s.

I know 2mp/s is alot but it seems like I'm not contributing much, by the end of the week should be back up around 3mp/s

[royanee]

Originally posted by vjs
I know 2mp/s is alot but it seems like I'm not contributing much, by the end of the week should be back up around 3mp/s

hehe, well at least you're contributing to the project. I had an interesting change in productivity. After a mobo swap with a friend (I had borrowed his Athlon XP 2800+), I was left with an Athlon XP 1600+... I went from 670 or so kp/s (the actual numbers are in the speed comparison thread) to about 400 kp/s (it was 360 kp/s, but I'm using the wine + Windows proth_sieve_cmov speed trick). It's so much slower, but not-too-surprisingly, much quieter and fewer stability issues.

[Keroberts1]

I was left with an Athlon XP 1600+... I went from 670 or so kp/s (the actual numbers are in the speed comparison thread) to about 400 kp/s (it was 360 kp/s, but I'm using the wine + Windows proth_sieve_cmov speed trick). It's so much slower, but not-too-surprisingly, much quieter and fewer stability issues.

This was a quote by someone in the co-ordination thread. What does this wine plus windows CMOV mean is this just to run CMov on linux? Or is it trick for use on windows machines that are less stable. I have a athalon 2000 that crashes pretty quickly when i start CMOV. The heat is the main issue but i haven't been able to get that takencare of yet so is there a way to run the siever at a slightly slower pace as to not create so much heat. I'm pretty sure its all of the floating point operations that are generating the highy temperatures. Its a little high when I'm not runing the siever but when I am it peaks around 88 degrees C. Yes i know that is bad so noone needs to tell me that.

[larsivi]

Originally posted by Keroberts1

This was a quote by someone in the co-ordination thread. What does this wine plus windows CMOV mean is this just to run CMov on linux? Or is it trick for use on windows machines that are less stable. I have a athalon 2000 that crashes pretty quickly when i start CMOV. The heat is the main issue but i haven't been able to get that takencare of yet so is there a way to run the siever at a slightly slower pace as to not create so much heat. I'm pretty sure its all of the floating point operations that are generating the highy temperatures. Its a little high when I'm not runing the siever but when I am it peaks around 88 degrees C. Yes i know that is bad so noone needs to tell me that.

I think the conclusion in an earlier discussion was that the windows binary is more efficiently compiled than the linux binary. This combined with the fact that linux is more efficient than windows meant that running the windows proth_sieve under Wine in linux was faster than running it under windows itself. The stability issue in question was about the hardware, not software.

[royanee]

Actually, that was me who posted that. It is faster in Linux to run the Windows binary under Wine, than to run the native Linux binary. (Though I haven't compared it to the statically linked Linux binary). I don't know how fast this machine goes on Windows, but it's probably faster, unless the terminal process in Windows is slow, which is very possible.

[Grafux]

Removed reservation /ceselb

Correct me if I'm wrong, but that should look like this:

Grafux:~ Grafux$ ./NbeGon_010_osx -s=SoB.dat -f=SoB.del -d=1.42 -p=630650G-631000G
N-be-gone v0.10sob (OSX) 52-bit 2003/01/24, Phil Carmody
# DLOG[10/1.420000] Huge (5160*3632)
# 2.6 p=630650000000077 (#f=0) hash overflows: 0|0|0|0|0|0
# 4.7 p=630650000048143 (#f=0) hash overflows: 4|0|0|0|0|0
# 7.4 p=630650000113669 (#f=0) hash overflows: 7|0|0|0|0|0
# 10.4 p=630650000179319 (#f=0) hash overflows: 11|0|0|0|0|0
# 13.5 p=630650000244743 (#f=0) hash overflows: 17|0|0|0|0|0
# 16.7 p=630650000310299 (#f=0) hash overflows: 26|0|0|0|0|0
# 20.0 p=630650000375873 (#f=0) hash overflows: 32|0|0|0|0|0
# 23.1 p=630650000441387 (#f=0) hash overflows: 40|0|0|0|0|0
# 26.2 p=630650000506909 (#f=0) hash overflows: 47|0|0|0|0|0
# 29.3 p=630650000572421 (#f=0) hash overflows: 53|0|0|0|0|0
# 32.5 p=630650000637971 (#f=0) hash overflows: 56|0|0|0|0|0

So the "G" tells the number to be "giga" or milllion? Is that right?

[vjs]

Looking good glad you got it figured out!!!

Your correct on the G and doublechecking your number

630650000637971

630650,000,637,971 (1G = 1,000,000,000) your good to go.

against one of my factors for comparison (My range 504000-513000)

630650000637971
504769849354361 | 10223*2^11749517+1

[Grafux]

"G" is billions, of course. You can tell I'm not a mathematician.

So I haven't found any factors/primes, unless I see a newly modified/created 'fact.txt' in the same folder as SoB.dat, right? Otherwise, NBeGon will finish sieving and produce no output if there are not significant findings in my range.

Hope I find some after all this!

[Mystwalker]

Originally posted by Grafux
So I haven't found any factors/primes, unless I see a newly modified/created 'fact.txt' in the same folder as SoB.dat, right? Otherwise, NBeGon will finish sieving and produce no output if there are not significant findings in my range.

In addition, NBeGon will increase the #f count (f for factor).

[Joe O]

If I'm not mistaken, the -f=SoB.del in the parameter list will cause NbeGon to write factors to the file SoB.del not fact.txt

[Mystwalker]

Wasn't it the other way round (SoB.del as default)?

[Joe O]

Originally posted by Mystwalker
Wasn't it the other way round (SoB.del as default)?

Yes, AFAIR SoB.del is the default.
I posted what I posted because in Grafux's first post he had explicitly set it to SoB.del, and in his next post he talked about fact.txt being modified/created. He would be a long time waiting for that, and might miss some factors.

[cedricvonck]

630600-630650 CedricVonck [complete]

2 factors:

630632836737929 | 4847*2^17664591+1
630647310534011 | 21181*2^10397828+1


I do not think that the sieve submit form did take my username into account.
Even if I had logged in.

[vjs]

Sometimes you have to

1. Close all iexplorer windows.
2. Login and check the remember me box.
3. Close all iexplorer again
4. Go directly to the page http://www.seventeenorbust.com/sieve

Now in the top left corner it will say logged in as ...

5. Submit your factors


A few of my computers show this behavior from time to time.

[Nuri]

Alternatively

1. login
2. go to sieve submission page
3. if your name does not appear on the upper left corner, simply refresh the page

it will appear..

[vjs]

Nuri,

Sometimes the simplest solution is the easiest, don't you hate cookies and iexplorer. Neither do what you want them to but broadcast PI to spammers .

[Silverfish]

chris stats say 49 factors after [complete], for the recently completed range. However, I made a copy of his results, and only 6 of them are unique factors (from fact.txt), which I gather is what should be listed after [complete]. The others were duplicates/excluded, or out of range factors.

[chris]

I only claimed factors as new as the verifying script told me (basically, i do not 'need' a single factor marked as new, BUT then the script has an error. Thats why I posted all the .txt files made by my most likely only sieve ever to be done by me).

FYI i post it again:

630500-630600 chris [complete] 49 factors found

As this is probably my first and last sieve (going to secondpass), i post the details :-)

fact.txt:

Factors
630502283108611|4847*2^15459183+1,630518226651589|
55459*2^4973386+1,630524523748213|10223*2^6076361+
1,630553039164781|24737*2^17095303+1,6305564570590
97|4847*2^13880991+1,630594205955369|55459*2^15377
554+1,
Verification Results
630502283108611 4847 15459183 verified. 630518226651589 55459 4973386 verified. 630524523748213 10223 6076361 verified. 630553039164781 24737 17095303 verified. 630556457059097 4847 13880991 verified. 630594205955369 55459 15377554 verified.

Factor table setup returned 1
Test table setup returned 1

6 of 6 verified in 0.32 secs.
6 of the results were new results and saved to the database.

factexcl.txt:

Factors
630501819547291|67607*2^1567299+1,630503020460129|
4847*2^12546255+1,630508296720077|67607*2^13350107
+1,630509258453093|10223*2^10718789+1,630513570390
383|4847*2^6485031+1,630513653497813|10223*2^18473
453+1,630516073868297|10223*2^7204493+1,6305179379
90383|19249*2^4328798+1,630518581425899|27653*2^92
28813+1,630518986879861|55459*2^7967962+1,63052218
8344579|55459*2^17492518+1,630523650958871|24737*2
^3453127+1,630529797283667|55459*2^2620798+1,63053
3550189827|21181*2^7338860+1,630533935843711|55459
*2^18557986+1,630536703592349|33661*2^13995144+1,6
30537711517319|33661*2^5097720+1,630541395940919|6
7607*2^15900571+1,630542285338183|27653*2^11689845
+1,630547089263687|33661*2^7170624+1,6305531315598
23|22699*2^10136206+1,630556063381957|10223*2^7741
985+1,630557910291493|67607*2^6952955+1,6305621713
18039|67607*2^7043747+1,630568702912541|33661*2^15
856800+1,630569569576691|24737*2^6590911+1,6305707
85620553|55459*2^15556186+1,630572376530153|55459*
2^3814522+1,630573985178651|67607*2^2287739+1,6305
74198756351|67607*2^17489339+1,630575029888477|102
23*2^18912665+1,630576293257283|55459*2^5503306+1,
630581350267969|33661*2^16621368+1,630581350267969
|33661*2^16621368+1,630581515253609|19249*2^174882
02+1,630581918276861|33661*2^7961568+1,63058306705
7381|10223*2^1385117+1,630583584835913|24737*2^401
9959+1,630583699871819|24737*2^19129543+1,63059540
4666527|10223*2^1831349+1,630597639368183|55459*2^
11977918+1,630599627143489|55459*2^2975818+1,63059
9656162337|67607*2^5539827+1,
Verification Results
630501819547291 67607 1567299 verified. 630503020460129 4847 12546255 verified. 630508296720077 67607 13350107 verified. 630509258453093 10223 10718789 verified. 630513570390383 4847 6485031 verified. 630513653497813 10223 18473453 verified. 630516073868297 10223 7204493 verified. 630517937990383 19249 4328798 verified. 630518581425899 27653 9228813 verified. 630518986879861 55459 7967962 verified. 630522188344579 55459 17492518 verified. 630523650958871 24737 3453127 verified. 630529797283667 55459 2620798 verified. 630533550189827 21181 7338860 verified. 630533935843711 55459 18557986 verified. 630536703592349 33661 13995144 verified. 630537711517319 33661 5097720 verified. 630541395940919 67607 15900571 verified. 630542285338183 27653 11689845 verified. 630547089263687 33661 7170624 verified. 630553131559823 22699 10136206 verified. 630556063381957 10223 7741985 verified. 630557910291493 67607 6952955 verified. 630562171318039 67607 7043747 verified. 630568702912541 33661 15856800 verified. 630569569576691 24737 6590911 verified. 630570785620553 55459 15556186 verified. 630572376530153 55459 3814522 verified. 630573985178651 67607 2287739 verified. 630574198756351 67607 17489339 verified. 630575029888477 10223 18912665 verified. 630576293257283 55459 5503306 verified. 630581350267969 33661 16621368 verified. 630581350267969 33661 16621368 verified. 630581515253609 19249 17488202 verified. 630581918276861 33661 7961568 verified. 630583067057381 10223 1385117 verified. 630583584835913 24737 4019959 verified. 630583699871819 24737 19129543 verified. 630595404666527 10223 1831349 verified. 630597639368183 55459 11977918 verified. 630599627143489 55459 2975818 verified. 630599656162337 67607 5539827 verified.

Factor table setup returned 1
Test table setup returned 1

43 of 43 verified in 1.19 secs.
42 of the results were new results and saved to the database.

factrange.txt:

Factors
630543670897061|4847*2^291423+1,630578348165369|67
607*2^22400699+1,
Verification Results
630543670897061 4847 291423 verified.

Factor table setup returned 1
Test table setup returned 1

1 of 2 verified in 0.03 secs.
1 of the results were new results and saved to the database.

You see, I do not 'claim' i found 49 new factors, I only BEEN TOLD I DID.

Cheers,

Chris

[Mystwalker]

The number that mainly matters is the amount of new factors in fact.txt...
But you're right - this is mentioned nowhere AFAIK.

[Grafux]

So my 'SoB.del' is still ... empty.

Can anyone estimate the time required to complete my range? NbeGon_010 has been running on this G4 Dual 867Mhz for 250 hours now, and I would like to reboot this machine in the near future.

Will quitting the process mean that I will have to start over?

[vjs]

What is your client speed?

I would spect that you find 1 factor every 20-25G,

also you should be able to restart no problem it will pickup where it left off..

There will be a file called SoB.bat

The contents should be something like

-s=SoB.dat -f=SoB.del -d=1.36 -p=630692865456137

The problem with sieve is you could go days without finding factors then get 3 or 4 in a row.

[Grafux]

The client speed is a dual processor G4 @867Mhz. That's roughly equivalent to a PIII @1200Mhz although I have no idea how optimized NBeGon is for the G4 architecture.

So my range must be somewhat large seeing as how it has taken over 10 days and counting?

[Mystwalker]

According to the start of the program you posted back some time, you're running at approx. 20kp/s.

This means that you do 25G in ~350 hours. I guess each G4 has a sieving client running, right? If yes, you should have done ~40G. So a factor seems to come late.

Nevertheless, I'd suggest finding a project where G4s scream. Then, you could look for someone with there suboptimal architectures (maybe P3/Athlon). Now, you offer him to exchange work, in a way that you both (and both projects, of course!) profit!

[vjs]

20 kp/s for a G4 does seems a little low, hopefully chucks new client will come out soon...

Mystwalker is correct if the G4 is only getting 20 kp/s another project would be a better solution. I know your an arsian like myself, check out the treads there... I believe TFY does work well with G4's, if this is the case kb9skw would probably be happy to do a swap. He is into both TFY and TPR-SoB, you could probably swap those 20 for 200 in a heart beat.

[Grafux]

I will look into swapping, thanks for the advice.

Incidentally I have one CPU sieving and one Folding, so I guess 20k it is.

[cjohnsto]

I was just wondering if some of the lower reserved ranges with no activity will be reassigned. These ranges are relatively rich with factors and some of the reservers have had no reported factors for months. Perhaps one of the coordinators should chase them up and see what is happening. As an example
alexr's last facor was submitted 02-Oct-2004. Did they complete the range? Still working on it? If not where did they abandon it?

[vjs]

cjohnsto,

By looking at his ranges "alexr" it looks like he did actually complete them he just hasn't reported back yet. Several people have tried to contact him in the past including myself. You might also consider that there are alot of other possibilites "holes" that should be looked at a little closer. These are from ranges that infact have been reported as complete.

Joe and myself have been testing some of these ranges with limited success, what seems to work best is keeping an eye on the factoring people. Frodo for example "a really aware factorer" will report when he finds a factor that should have been found through sieving. I will generally go back and test that range fully.

Please do not retest any ranges, you may infact be tripple or even quadruple checking ranges. At present our best bet is to continue working on ranges that we know for certain havn't been tested yet. To create a double check sieve effort at this time would be a real nightmare for all involved.

Search around in the other treads regarding missed factors, double check sieve, high and low range sieving, etc... You'll see this has come up in the past.

[cjohnsto]

I did not intend on starting these ranges again, merely wondering about thier status.

[vjs]

Status...

Bascially most everything less than 26000 is either completed or being worked on.
Some of them have even been tripple checked.

As for the higher ranges? Time will tell but currently retesting those is not the thing to do without some careful consideration or proof of missed factors. There are a few above 200000 but they have lower success rates "guesses about missed factors or incomplete ranges declared complete" than those at lower p.

[Nuri]

looking at alexr's stats page, it seems reasonable to assume they're completed.



even if not so, it really is not worth resieving those four ranges of him.

it's only a matter of bookkeeping. may be we can move his ranges to archive with a tiny note attached, like:

359000-360000 alexr [not reported as complete yet]
361050-361200 alexr [not reported as complete yet]
370000-372000 alexr [not reported as complete yet]
374000-374500 alexr [not reported as complete yet]