DC server how should we proceed

ltd · 2008-07-29, 16:26

As I think everybody helping to push the project should have a chance to decide about our future strategie I want to start a discussion about the future strategie for the feeding of the DC server.

At the moment I see two possible strategies:

1. Hand out one k to slowly catch up with first pass and every time there is a pattern of errors insert all pairs of that pattern before moving forward.

2. Hand out one k and ignore error pattern until we have reached first pass level. When we have the big picture come back and clean up all the error patterns depending on their priority.

If somebody has other ideas or reasons why we should prever one strategie please respond.

VJS · 2008-07-29, 20:22

Lars,

First off I think your doing a great job with the DC server, your attention to patterns and errors is fantastic.

I would personally vote for #1.

But in all honesty I would actually make a #3.

3. Hand out one k to slowly catch up with first pass. When Lars see's any error pattern or has a hunch on a particular set of tests. Let Lars decide if it's worth the switch... switch.

-------------------

I would also like to comment on the inserted tests. If you see a pattern its worth switching over for several reasons.

First the only risk is an early doublecheck and a slower progression of the test k.

Second when this group of tests are added we will either find errors (the major purpose of the DC que) or have a better understanding of the error rate at that particular n level. In either case this is win win.

Third I would rather do a slower progess of one k with the DC server than have to go back. I somewhat believe that doublechecking one k upto the firstpass level is sufficient with doublechecking. In other words if the error rates look good with the current approach I see no need to run doublechecks on all k's considering our firstpass level and the possibility of removing serveral k's through firstpass testing eliminating those doublechecks all together. Consider the last prime, it removed alot of double checks from the system... ~15%???

japelprime · 2008-08-01, 11:21

Are all # of k even sensitive in the same n range ?

ltd · 2008-08-01, 14:40

As far as we know from DC testing everything below n=1.45M there is no difference between the k values.

k=168451 was choosen due to the fact that it it has a low number of special range reservations which gives us a good picture about error rates over the complete test range.

2008-07-29, 20:22	#2
VJS Dec 2004 100101011₂ Posts	Lars, First off I think your doing a great job with the DC server, your attention to patterns and errors is fantastic. I would personally vote for #1. But in all honesty I would actually make a #3. 3. Hand out one k to slowly catch up with first pass. When Lars see's any error pattern or has a hunch on a particular set of tests. Let Lars decide if it's worth the switch... switch. ------------------- I would also like to comment on the inserted tests. If you see a pattern its worth switching over for several reasons. First the only risk is an early doublecheck and a slower progression of the test k. Second when this group of tests are added we will either find errors (the major purpose of the DC que) or have a better understanding of the error rate at that particular n level. In either case this is win win. Third I would rather do a slower progess of one k with the DC server than have to go back. I somewhat believe that doublechecking one k upto the firstpass level is sufficient with doublechecking. In other words if the error rates look good with the current approach I see no need to run doublechecks on all k's considering our firstpass level and the possibility of removing serveral k's through firstpass testing eliminating those doublechecks all together. Consider the last prime, it removed alot of double checks from the system... ~15%??? Last fiddled with by VJS on 2008-07-29 at 20:24

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Issue with gnfs failing to proceed.	Greebley	Aliquot Sequences	4	2014-04-17 17:31
New server!	Xyzzy	Lounge	8	2005-08-29 15:30
New ECM-server available	andi314	Factoring	3	2003-08-31 11:22
46-46.1M to 2^59. Will proceed to 47M	lycorn	Lone Mersenne Hunters	2	2003-03-10 10:10
New Server Hardware and price quotes, Funding the server	Angular	PrimeNet	32	2002-12-09 01:12

2008-07-29, 16:26	#1
ltd Apr 2003 2²·193 Posts	DC server how should we proceed As I think everybody helping to push the project should have a chance to decide about our future strategie I want to start a discussion about the future strategie for the feeding of the DC server. At the moment I see two possible strategies: 1. Hand out one k to slowly catch up with first pass and every time there is a pattern of errors insert all pairs of that pattern before moving forward. 2. Hand out one k and ignore error pattern until we have reached first pass level. When we have the big picture come back and clean up all the error patterns depending on their priority. If somebody has other ideas or reasons why we should prever one strategie please respond. Last fiddled with by ltd on 2008-07-29 at 16:27

2008-08-01, 11:21	#3
japelprime "Erling B." Dec 2005 130₈ Posts	Are all # of k even sensitive in the same n range ?

2008-08-01, 14:40	#4
ltd Apr 2003 1100000100₂ Posts	As far as we know from DC testing everything below n=1.45M there is no difference between the k values. k=168451 was choosen due to the fact that it it has a low number of special range reservations which gives us a good picture about error rates over the complete test range.