Two milestones in a week

Batalov · 2008-07-08, 01:46

July 4, 2008: All exponents below M(20996011) tested at least once.
July 8, 2008: All exponents below 17,000,000 double-checked.

Brian-E · 2008-07-10, 16:05

Quote:

Originally Posted by Batalov

July 4, 2008: All exponents below M(20996011) tested at least once.
July 8, 2008: All exponents below 17,000,000 double-checked.

No doubt correct and very pleasing to know! I guess the data I get from the PrimeNet 4.0 server which gives the following tests of exponents under 17000000 in progress...

Code:

16320053 D   66   3911227   505.3  19.8  54.8  07-Jun-08 11:00  21-Feb-07 07:30  S587032        C6FC3A1B9
16523869     66  15225697    27.2 -22.2  37.8                   13-Jun-08 10:58  drno           intoe
16809791 D   66  12857152   597.1  -4.3  55.8  15-Jun-08 09:55  21-Nov-06 12:43  cucciolo       C47E0FFBF
16855673     66   8085829    29.2  72.7  52.7  30-Jun-08 08:13  11-Jun-08 11:31  mum            drm-Susi
16860589     66   3798087     4.9  83.1  83.1                   05-Jul-08 17:11  jangross       Pentium_Home
16863859     66              47.6 -37.6  22.4                   24-May-08 00:35  S503730        C91F878D0
16874471     66     84068    30.4 -13.4  46.6                   10-Jun-08 06:33  S508035        CC1444F86
16903207 D   66  11498816   472.3   4.8  64.8  09-Jul-08 09:06  26-Mar-07 08:20  S232665        flormultstuf
16928011     66     12799    17.8  -9.8  50.2                   22-Jun-08 19:35  S476738        CDEC58442
16955867     66  16181568    39.8 -38.8  21.2                   31-May-08 19:54  S450833        CEAE9C5C4
16959469     66              19.0 -11.0  49.0                   21-Jun-08 14:53  S450990        CC0FCCF70

...plus many other first-time tests below exponent 20996011 is superseded by tests which have been completed on version 5? Or is there some other explanation?

Batalov · 2008-07-10, 17:02

They are both correct.

The way you found them is the same way I found them, then double-checked the V5 server's reports for the three exponents - they were available for LL-D. I had some spare cycles over the 4th of July weekend and manually added the three eyesores to the worktodo.txt on a box that was used for 800M-1B low-lever factoring.

Don't worry about the credit - the V5 credit is a 'sandbox' credit (and even whatever it is - I gladly agree to disown it), while the V4 credit is the real one. If the exponents' V4 owners finish them, they will receive full credit. But it is a bit silly to do a 4-cpu day job over almost two years; they seem to be more likely to never finish. (As far as I've heard if the jobs reach two years, they will have expired, but it's a bit of a wait.)

So, in a way this milestone is the slight of a hand.

Serge

Brian-E · 2008-07-10, 18:52

Yes, I see. I suppose the two parallel servers V4 and V5 make a useful way of keeping the milestones moving whilst avoiding very obviously poaching the jobs of people who use slow computers or whose computers rarely run but who are nevertheless dedicated contributors to GIMPS. I hadn't thought of it that way before. Thanks for your explanation.

Batalov · 2008-07-10, 19:28

Guilty as charged. I admit to poaching the three exponents for an early LL-T ("triple-check") (while the double-check is pending) without credit (with toy credit).

On a plus side, a) I tested my hardware (one cannot test it by running the first-checks!) - the residues matched for all three P's. b) I tested the V5 server and the p64v256.zip Windows-64bit binary client (my other client programs are all different). c) The owners will receive full credit when they finish and there's no chance of a M45 there.

Now, a mental exercise to the reader - think about the other milestone. It's in the initial message not incidentally... In the same light, that is some serious global poaching. 970 first-time tests, no less.

Even more seriously, (most of) it is actually not! I've had enough exponents under my belt to see the following scenario happening -
1. The original owner loses the V4 assignment (60 days without check-ins)
2. Someone else gets it from the V4 server.
3. ...finishes it!
4. ...the original owner's computer connects to the V4 server after a long-ish sleep and re-reserves the exponent (!!) with a new date (mostly with a 2008 date, this time)

==> Hence the effect that you see in the snippet of the table that is conveniently in your message...

Here's a real-life example of this happening with my perfectly legally obtained (and long ago finished) exponents 35976221 and 35982629. They are in the status.txt again!! Don't tell me that I poached those.

Serge

Brian-E · 2008-07-10, 20:38

Yep, after a few checks on http://v5www.mersenne.org I've convinced myself that what you say is correct. Obviously the status file of current jobs on the V4 server is not the reliable guide to progress of the GIMPS project that I thought it was.

After thinking about it, I don't personally think that there is anything wrong with the phenomenon of accepting the re-reserving of the exponents from machines which have "woken up" after someone else has completed their tests. It seems to me the right thing to do in those circumstances. Their tests are, after all, in some degree of advanced state, and if and when they eventually produce results those will be recorded as a double (or triple) check. The confusing part is that we don't see this double-check status in the V4 status file.

garo · 2008-07-11, 10:01

These spurious triple-checks are regularly removed from the v4 server but they keep coming back.

As far as long over-due exponents are concerned, in v5 small exponents that are stopping a milestone will only be given to computers that have a track record of finishing exponents fast.

2008-07-08, 01:46	#1
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 10010100000100₂ Posts	Two milestones in a week July 4, 2008: All exponents below M(20996011) tested at least once. July 8, 2008: All exponents below 17,000,000 double-checked. Attached Thumbnails Last fiddled with by Batalov on 2008-07-08 at 02:08

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Computer starts cackling twice a week after two years OK	fivemack	Hardware	11	2015-05-09 00:52
Grab this chance for a rare sight this week	cheesehead	Astronomy	10	2011-03-14 22:53
slow news week?	ixfd64	Lounge	11	2008-09-29 00:02
Only emergency updates this week	ltd	Prime Sierpinski Project	4	2007-02-09 02:21
Hardware Of the Week #1	moo	Hardware	4	2005-10-19 15:58

2008-07-10, 17:02	#3
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 2²×23×103 Posts	They are both correct. The way you found them is the same way I found them, then double-checked the V5 server's reports for the three exponents - they were available for LL-D. I had some spare cycles over the 4th of July weekend and manually added the three eyesores to the worktodo.txt on a box that was used for 800M-1B low-lever factoring. Don't worry about the credit - the V5 credit is a 'sandbox' credit (and even whatever it is - I gladly agree to disown it), while the V4 credit is the real one. If the exponents' V4 owners finish them, they will receive full credit. But it is a bit silly to do a 4-cpu day job over almost two years; they seem to be more likely to never finish. (As far as I've heard if the jobs reach two years, they will have expired, but it's a bit of a wait.) So, in a way this milestone is the slight of a hand. Serge

2008-07-10, 18:52	#4
Brian-E "Brian" Jul 2007 The Netherlands 110011000101₂ Posts	Yes, I see. I suppose the two parallel servers V4 and V5 make a useful way of keeping the milestones moving whilst avoiding very obviously poaching the jobs of people who use slow computers or whose computers rarely run but who are nevertheless dedicated contributors to GIMPS. I hadn't thought of it that way before. Thanks for your explanation.

2008-07-10, 19:28	#5
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 2²·23·103 Posts	Guilty as charged. I admit to poaching the three exponents for an early LL-T ("triple-check") (while the double-check is pending) without credit (with toy credit). On a plus side, a) I tested my hardware (one cannot test it by running the first-checks!) - the residues matched for all three P's. b) I tested the V5 server and the p64v256.zip Windows-64bit binary client (my other client programs are all different). c) The owners will receive full credit when they finish and there's no chance of a M45 there. Now, a mental exercise to the reader - think about the *other* milestone. It's in the initial message not incidentally... In the same light, *that* is some serious global poaching. *970* first-time tests, no less. Even more seriously, (most of) it is actually not! I've had enough exponents under my belt to see the following scenario happening - 1. The original owner loses the V4 assignment (60 days without check-ins) 2. Someone else gets it from the V4 server. 3. ...finishes it! 4. ...the original owner's computer connects to the V4 server after a long-ish sleep and re-reserves the exponent (!!) with a new date (mostly with a 2008 date, this time) ==> Hence the effect that you see in the snippet of the table that is conveniently in your message... Here's a real-life example of this happening with my perfectly legally obtained (and long ago finished) exponents 35976221 and 35982629. They are in the status.txt again!! Don't tell me that I poached those. Serge

2008-07-10, 20:38	#6
Brian-E "Brian" Jul 2007 The Netherlands 7·467 Posts	Yep, after a few checks on http://v5www.mersenne.org I've convinced myself that what you say is correct. Obviously the status file of current jobs on the V4 server is not the reliable guide to progress of the GIMPS project that I thought it was. After thinking about it, I don't personally think that there is anything wrong with the phenomenon of accepting the re-reserving of the exponents from machines which have "woken up" after someone else has completed their tests. It seems to me the right thing to do in those circumstances. Their tests are, after all, in some degree of advanced state, and if and when they eventually produce results those will be recorded as a double (or triple) check. The confusing part is that we don't see this double-check status in the V4 status file.

2008-07-11, 10:01	#7
garo Aug 2002 Termonfeckin, IE 101011001100₂ Posts	These spurious triple-checks are regularly removed from the v4 server but they keep coming back. As far as long over-due exponents are concerned, in v5 small exponents that are stopping a milestone will only be given to computers that have a track record of finishing exponents fast.