Server behavior for doing quick double-checks if error code != 0

[GP2]

Is anyone familiar with the exact behavior of the server?


I believe the server does the following:

When a result is returned with nonzero error code, the server immediately makes that exponent available for a second "first-time" LL test. This is done to quickly catch bad results without waiting years for the leading edge of regular double checks to catch up.

My question is, is this done for all error codes that are nonzero, or are some error codes like 0x000x00 also skipped over? Also, approximately when was this server behavior implemented?

Thanks.

[NickGlover]

I don't believe the server automatically handles any aspect of the double-checking process. Every once in a while, George manually re-releases the exponents that had too many errors. I do not know the precise error codes that George re-releases, but I do know the 0X000X00 codes are not re-released.

[Prime95]

Nick is right, I do this manually (actually I run an SQL query).

0X000X00 are not rerun.
00XX0000 are not rerun.

[GP2]

Originally posted by Prime95
Nick is right, I do this manually (actually I run an SQL query).

0X000X00 are not rerun.
00XX0000 are not rerun.

Just to make sure I'm understanding this:

Does this mean


(errorcode & ~0x0F000F00) == 0 is not rerun
(errorcode & ~0x00FF0000) == 0 is not rerun


or do the two X's have to be identical


01000100
02000200

etc.

and

00110000
00220000

etc.

[dsouza123]

A standard usage is X means, doesn't matter what the value is, so as long as the error code only has non zeros where an X is it is not rerun.

00000000 are not rerun.

00000X00 are not rerun.
0X000000 are not rerun.
0X000X00 are not rerun.

000X0000 are not rerun.
00X00000 are not rerun.
00XX0000 are not rerun.

[Prime95]

Originally posted by dsouza123
A standard usage is X means, doesn't matter what the value is

I'm guilty of non-standard usage

XY00XY00 is not rerun if the XY values are the same. That is, 03000300 is not rerun, but 03000400 is rerun.

00XX0000 is not rerun for any values of XX.

[Xyzzy]

This probably is a bit off topic, but I've always wondered what the stuff in results.txt means... Like this:

WZ1: 82E972AA,4529786,00000000

I think the "WZ1" is the program code, and the next part is like a CRC checksum or something, and the last part is the error code...

Can someone break it all down into crazy detail?

[NickGlover]

From "readme.txt":

M1992031 is not prime. Res64: 6549369F4962ADE0. WV1: B253EF24,1414032,00000000
	This means 2^1992031-1 is not prime - a Lucas-Lehmer test says so.
	The last 64 bits of the last number in the Lucas-Lehmer sequence
	is 6549369F4962ADE0.  At some future date, another person will verify
	this 64-bit result by rerunning the Lucas-Lehmer test.  WV1 is the
	program	version number.  B253EF24 is a checksum to guard against email
	transmission errors.  1414032 can be ignored it is used as part
	of the double-checking process.  The final 00000000 value is a set
	of 4 counters.  These count the number of errors that occurred during
	the Lucas-Lehmer test.

I will add that 1414032 from this example is the shift count. I believe it helps ensure there are no reproducible bugs causing multiple tests to generate the same incorrect residue (i.e. from a hardware or software bug).

I don't know exactly what each of the 4 error counters represent. Maybe George can answer.

[eepiccolo]

Originally posted by NickGlover
I don't know exactly what each of the 4 error counters represent. Maybe George can answer.

I'm sure I've seen it in another thread, but I can't remember where.

[eepiccolo]

Originally posted by eepiccolo
I'm sure I've seen it in another thread, but I can't remember where.

Never mind, the thread I was thinking of was this, but it doesn't have that info.

[apocalypse]

There is some discusion of error codes in this thread