error rates

drakkar67 · 2005-11-26, 14:26

with the available data for first pass and second pass residuals do you have an error rate?

AFAIK SoB has a high error rate of about 5%. The last three primes found in PSP have big gaps between them. Are you suspicious about a prime hiding there?

drakkar67

ltd · 2005-11-26, 14:46

Error rates are very low so far.
There is only one slot where we have higher error rates and that is a range where we had results from a bad llr Client version. This range is doublechecked by Mytwalker.

For the errors found so far:
This is an information i put on the forum in the end of october:

Here is a breakdown of the different numbers of tests done so far.

We have:
118165 first pass PRP test
22797 second pass test
59 third pass tests due to not fitting results

These 59 tests split into the following groups:

3 tests come from double reservations where one of the tests is run with
PRP and the other with llr version >= 3.5 Therefore there were incompatible residues.

50 tests are from the range of tests where results came back from a buggy llr client

6 tests are real errors.

I personaly think that we have not missed a prime so far but i also think that there is one more before we reach n=2M.

Lars

VJS · 2005-12-07, 16:59

I'm sure this thread is related to SoB so let me say that error rates are more of a per unit time bases. Therefore as tests take longer to complete your possibility of error increase. With n's less than 3M you don't have too much to worry about unless there are specific instances like the bad clients or users. Cudos on rechecking quickly.

Personally I wouldn't consider any doublechecking of tests until you reach about 3M... You'll probably eliminate a couple more by then, At that time I'd bring secondpass up to a level where you can test 1 first pass test in the time it takes to test 12 secondpass tests. From there on out keep the Time secondpass x 12 = Time firstpass

Some my argue higher or lower but 1 in 12, that should cover the potential 5-8% error rate as tests increase. Also you don't get too far ahead of yourself either.

drakkar67 · 2005-12-08, 06:09

do you know when we will pass the 192 K FFT length? and what will be the speed of clients then?
drakkar

Mystwalker · 2005-12-08, 10:41

I had to use 192 KB FFTs when I tested the n=1.725M candidates of the LLRnet server.
Unfortunately, I have no comparable speed data...

ltd · 2005-12-08, 17:20

The step from FFT 192K to FFT zero padded 224K happens somewhere between 2.12M and 2.14M. I have not tested the speed yet.
(Tests done with k=214519)

Lars

Mystwalker · 2005-12-08, 18:24

Just for clarification:

drakkar, did you mean "pass the limit to 192 KB FFT" or "pass to limit from 192 KB FFT to the next size"?

drakkar67 · 2005-12-09, 04:01

Quote:

drakkar, did you mean "pass the limit to 192 KB FFT" or "pass to limit from 192 KB FFT to the next size"?

I meant from 192 K to the next size.
Unfortunately it's not too far away. I hope we'll hunt a prime until 224K.

Citrix · 2005-12-11, 19:02

What are the new error rates with the double check work recently done?

Citrix

ltd · 2005-12-11, 19:46

Still no change in the error patern.
There are only new errors in the range from the buggy llr client.

Lars

Citrix · 2005-12-11, 20:26

Quote:

Originally Posted by ltd

There are only new errors in the range from the buggy llr client.

Lars

How many new ones?

2005-11-26, 14:26	#1
drakkar67 Feb 2003 Istanbul 2²·13 Posts	error rates with the available data for first pass and second pass residuals do you have an error rate? AFAIK SoB has a high error rate of about 5%. The last three primes found in PSP have big gaps between them. Are you suspicious about a prime hiding there? drakkar67

2005-12-07, 16:59	#3
VJS Dec 2004 13×23 Posts	I'm sure this thread is related to SoB so let me say that error rates are more of a per unit time bases. Therefore as tests take longer to complete your possibility of error increase. With n's less than 3M you don't have too much to worry about unless there are specific instances like the bad clients or users. Cudos on rechecking quickly. Personally I wouldn't consider any doublechecking of tests until you reach about 3M... You'll probably eliminate a couple more by then, At that time I'd bring secondpass up to a level where you can test 1 first pass test in the time it takes to test 12 secondpass tests. From there on out keep the Time secondpass x 12 = Time firstpass Some my argue higher or lower but 1 in 12, that should cover the potential 5-8% error rate as tests increase. Also you don't get too far ahead of yourself either. Last fiddled with by VJS on 2005-12-07 at 17:00

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Electrical Service Rates	storm5510	Hardware	178	2023-04-13 00:51
error rates and P-1 test	drakkar67	Prime Sierpinski Project	9	2008-05-26 14:29
LL Test Rates and GIMPS Promotion	Primenut	Lounge	14	2003-06-09 09:32
Error Rates	Prime95	Math	31	2002-09-06 14:34
Error rates revealed	Prime95	Math	1	2002-09-01 00:10

2005-11-26, 14:46	#2
ltd Apr 2003 2²·193 Posts	Error rates are very low so far. There is only one slot where we have higher error rates and that is a range where we had results from a bad llr Client version. This range is doublechecked by Mytwalker. For the errors found so far: This is an information i put on the forum in the end of october: Here is a breakdown of the different numbers of tests done so far. We have: 118165 first pass PRP test 22797 second pass test 59 third pass tests due to not fitting results These 59 tests split into the following groups: 3 tests come from double reservations where one of the tests is run with PRP and the other with llr version >= 3.5 Therefore there were incompatible residues. 50 tests are from the range of tests where results came back from a buggy llr client 6 tests are real errors. I personaly think that we have not missed a prime so far but i also think that there is one more before we reach n=2M. Lars

2005-12-08, 06:09	#4
drakkar67 Feb 2003 Istanbul 2²·13 Posts	do you know when we will pass the 192 K FFT length? and what will be the speed of clients then? drakkar

2005-12-08, 10:41	#5
Mystwalker Jul 2004 Potsdam, Germany 3·277 Posts	I had to use 192 KB FFTs when I tested the n=1.725M candidates of the LLRnet server. Unfortunately, I have no comparable speed data...

2005-12-08, 17:20	#6
ltd Apr 2003 1100000100₂ Posts	The step from FFT 192K to FFT zero padded 224K happens somewhere between 2.12M and 2.14M. I have not tested the speed yet. (Tests done with k=214519) Lars

2005-12-08, 18:24	#7
Mystwalker Jul 2004 Potsdam, Germany 1100111111₂ Posts	Just for clarification: drakkar, did you mean "pass the limit to 192 KB FFT" or "pass to limit from 192 KB FFT to the next size"?

2005-12-11, 19:02	#9
Citrix Jun 2003 2²·11·37 Posts	What are the new error rates with the double check work recently done? Citrix

2005-12-11, 19:46	#10
ltd Apr 2003 2²×193 Posts	Still no change in the error patern. There are only new errors in the range from the buggy llr client. Lars