View Single Post
Old 2004-03-18, 20:35   #1
nfortino
 
nfortino's Avatar
 
Nov 2003

3×5×11 Posts
Default Tracking the November 2003 release of error-prone exponents

Here is the status of the November release of 1144 error-prone exponents originaly metioned in this thread. http://www.mersenneforum.org/showthread.php?t=1381

1144 exponents were released

1095 of these had been tested once and needed a 2nd test:
785 had a harmless error code, but were returned by error prone machines
310 had a harmful error code.

49 exponents were tested twice and needed a 3rd test (both
original tests were by error-prone machines or had a harmful error code).

The status as of 2004-Mar-18 20:00 UTC:


Of the 49 (out of 1144) exponents released for a 3rd test
(because both original tests were error-prone),
18 are still pending,
1 needs a quintuple check,
3 need quadruple checks,
3 have expired and not been reassigned,
24 now have two matching results.

3 of these 24 have only good results,
21 of these 24 have results known to be bad.

The fraction of exponents in which both of the original tests are bad is
(assuming the original tests are at fault for mismatches) (1+3)/(1+3+24)=.143
Since the criterion for finding these exponents was an expected error rate of
.333 per test, this is about in-line with the expected .111

Of the 785 (out of 1144) exponents which had been released for a second test,
261 are still pending,
8 have been factored,
2 have apparently expired and not been reassigned,
4 have very recently cleared,
441 have been confirmed good,
and the remaining 69 have erroneous results.
Of these 69,
2 have two good results and one bad result,
67 still require a triple check.

This is an error rate of 69/(69+441)=.135,
far below the 0.333 error rate which was the criterion for detecting
error prone machines.


Finally, of the 310 (out of 1144) exponents released for a second check due to
harmfull error codes,
131 are still pending,
0 have been factored,
1 was recently cleared,
41 are now confirmed good,
the remaining 137 have erroneous results.
Of the 137,
2 have two good results and one bad result,
134 now need a triple check,
1 now needs a quadruple check.

This is an error rate of (1+136)/(1+136+41)=.770

More detail in the attached file:
Attached Files
File Type: zip November_batch_status_31804.zip (36.0 KB, 231 views)
nfortino is offline   Reply With Quote