mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Data

Reply
 
Thread Tools
Old 2012-03-12, 11:52   #12
retina
Undefined
 
retina's Avatar
 
"The unspeakable one"
Jun 2006
My evil lair

10100101110002 Posts
Default

Semi-serious suggestion: Have P95 ask the user at start-up if the machine has ECC RAM. Give three options: "Yes", "No" and "Don't know". The last two options are the same and mean no-ECC present. And an additional flag set if the user selected "don't know" to assume that they have no idea what RAM timings are or how to make them optimal for stability.

So then if there is no ECC and the user selects any LL assignment that is expected to take more than one month to complete then very strongly discourage such a test on the basis that it would most probably end up with a useless result. If the user chooses to override the suggestion and runs anyway then so be it and let them deal with whatever happens.

And perhaps additionally if the user selected "don't know", to discourage tests taking longer than two weeks.

Anyhow the time frames can be adjusted if they seem overly draconian but I would encourage something like this be implemented to help reduce wasted time and effort.
retina is offline   Reply With Quote
Old 2012-03-12, 12:46   #13
KyleAskine
 
KyleAskine's Avatar
 
Oct 2011
Maryland

29010 Posts
Default

Quote:
Originally Posted by retina View Post
Semi-serious suggestion: Have P95 ask the user at start-up if the machine has ECC RAM. Give three options: "Yes", "No" and "Don't know". The last two options are the same and mean no-ECC present. And an additional flag set if the user selected "don't know" to assume that they have no idea what RAM timings are or how to make them optimal for stability.

So then if there is no ECC and the user selects any LL assignment that is expected to take more than one month to complete then very strongly discourage such a test on the basis that it would most probably end up with a useless result. If the user chooses to override the suggestion and runs anyway then so be it and let them deal with whatever happens.

And perhaps additionally if the user selected "don't know", to discourage tests taking longer than two weeks.

Anyhow the time frames can be adjusted if they seem overly draconian but I would encourage something like this be implemented to help reduce wasted time and effort.
This cannot be a good idea. There are plenty of still contributing systems that might not be able to do a DC in two weeks. Never mind an LL test.

And the fewer questions, the better. Don't want to intimidate people right out of the gate.

Now, if there was some test selected that would take more than six months, then a pop-up warning might not be out of the question. But that is as aggressive as I would get.
KyleAskine is offline   Reply With Quote
Old 2012-03-12, 12:51   #14
cheesehead
 
cheesehead's Avatar
 
"Richard B. Woods"
Aug 2002
Wisconsin USA

170148 Posts
Default

Quote:
Originally Posted by KyleAskine View Post
Now, if there was some test selected that would take more than six months, then a pop-up warning might not be out of the question. But that is as aggressive as I would get.
So we have a second suggestion about time frame.

Could someone do some statistics showing numbers of errors versus elapsed time of assignment completion versus machine type, for various types of assignments?
cheesehead is offline   Reply With Quote
Old 2012-03-12, 13:50   #15
kladner
 
kladner's Avatar
 
"Kieren, ktony"
Jul 2011

3×19×167 Posts
Default

Quote:
Originally Posted by cheesehead View Post
.......Could someone do some statistics showing numbers of errors versus elapsed time of assignment completion versus machine type, for various types of assignments?
I don't know how to go about the study, but I would certainly be interested in the results, being that I run non-ECC RAM.
kladner is offline   Reply With Quote
Old 2012-03-12, 17:41   #16
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(3,3^1118781+1)/3

2·4,493 Posts
Default

Quote:
Originally Posted by lycorn View Post
More informatively - LL result (with error code). Just like you said, as expected.
Batalov is offline   Reply With Quote
Old 2012-03-12, 17:43   #17
Dubslow
Basketry That Evening!
 
Dubslow's Avatar
 
"Bunslow the Bold"
Jun 2011
40<A<43 -89<O<-88

3·29·83 Posts
Default

At least it's not all FFs.
Dubslow is offline   Reply With Quote
Old 2012-03-12, 21:11   #18
Brian-E
 
Brian-E's Avatar
 
"Brian"
Jul 2007
The Netherlands

2·23·71 Posts
Default

Quote:
Originally Posted by retina View Post
Semi-serious suggestion: Have P95 ask the user at start-up if the machine has ECC RAM. Give three options: "Yes", "No" and "Don't know". The last two options are the same and mean no-ECC present. And an additional flag set if the user selected "don't know" to assume that they have no idea what RAM timings are or how to make them optimal for stability.

So then if there is no ECC and the user selects any LL assignment that is expected to take more than one month to complete then very strongly discourage such a test on the basis that it would most probably end up with a useless result. If the user chooses to override the suggestion and runs anyway then so be it and let them deal with whatever happens.

And perhaps additionally if the user selected "don't know", to discourage tests taking longer than two weeks.

Anyhow the time frames can be adjusted if they seem overly draconian but I would encourage something like this be implemented to help reduce wasted time and effort.
My answer to the question about ECC memory would be "don't know" because I know nothing about computer hardware and can't be bothered to Google and find out what ECC means.

Since starting my participation in the project nearly 5 years ago I have completed 3 first-time LL tests and 18 double checks. The double checks all took between 1 month and 3 months to complete. The 3 first-time tests, all just below or above exponent 40 million since I did these some years ago, each took about 6 months.

Clearly by your criteria I would be strongly discouraged from taking part.

I would like to mention that my 18 double-checks have all, without exception, produced correct results insofar as they have agreed with a previous test. My first-time tests have not been double-checked and therefore their correctness is unknown: their error codes were however all zero.
Brian-E is offline   Reply With Quote
Old 2012-03-12, 21:23   #19
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(3,3^1118781+1)/3

2×4,493 Posts
Default

Quote:
Originally Posted by retina View Post
Semi-serious suggestion: Have P95 ask the user at start-up if the machine has ECC RAM. Give three options: "Yes", "No" and "Don't know". The last two options are the same and mean no-ECC present. And an additional flag set if the user selected "don't know" to assume that they have no idea what RAM timings are or how to make them optimal for stability.
It is a great suggestion, but no need to ask them, really. There's an API to get the answer from the computer.
Batalov is offline   Reply With Quote
Old 2012-03-12, 21:34   #20
axn
 
axn's Avatar
 
Jun 2003

3·1,511 Posts
Default

Quote:
Originally Posted by Batalov View Post
There's an API to get the answer from the computer.
There is an even easier method to get the answer: "No". ECC is a luxury enjoyed by servers (and that too, not all servers). From GIMPS context, "No" is the only answer that makes sense.
axn is offline   Reply With Quote
Old 2012-03-12, 22:16   #21
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(3,3^1118781+1)/3

2×4,493 Posts
Default

This is very true on many levels. The computer on which I type this message has ECC, and I can verify it (and P95 would be able to). But I cannot run anything on it (legally! physically, yes, and I have run some untill 2008 when I got a clear message from management to cease along the line "We're going to have this conversation this once only; next time ... you get the idea").

When I saw early that someone's M3xx,xxx,xxx top runner's computer was a Xeon workstation not very dissimilar from what I have, I though - well here's a lucky guy who has a chance to do it right (and hopefully does it with a permission from his management). There's probably a tiniest segment of ECC users; is it worth bothering about special treatment for them, that's an open question.
Batalov is offline   Reply With Quote
Old 2012-03-12, 22:20   #22
Mini-Geek
Account Deleted
 
Mini-Geek's Avatar
 
"Tim Sorbera"
Aug 2006
San Antonio, TX USA

17×251 Posts
Default

Quote:
Originally Posted by axn View Post
There is an even easier method to get the answer: "No". ECC is a luxury enjoyed by servers (and that too, not all servers). From GIMPS context, "No" is the only answer that makes sense.
As a generality? Sure. In the fringe cases, though, the answer is indeed "Yes". I'd be in favor of: if a user selects a >6 month test (or perhaps simply "a 100M digit test") and the software can see that the memory is not ECC (or the reservation is manual, so that PrimeNet cannot know at reservation time whether ECC is present), a warning pops up informing them of the situation, and giving them the option to change their work assignment or go ahead with it.
Mini-Geek is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Number of LLR tests completed so far? ellipse Twin Prime Search 26 2019-09-28 17:19
Largest number of LL tests before matching residues achieved? sdbardwick Lounge 1 2015-02-03 15:03
Completed 29M work not showing as completed in GPU72 Chuck GPU to 72 2 2013-02-02 03:25
Largest LL Test Ever Completed jinydu Lounge 40 2010-03-22 20:54
need Pentium 4s for 5th largest prime search (largest proth) wfgarnett3 Lounge 7 2002-11-25 06:34

All times are UTC. The time now is 00:46.

Thu Apr 2 00:46:37 UTC 2020 up 7 days, 22:19, 3 users, load averages: 0.87, 1.06, 1.15

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.