dealing with abandoned tests

[Keroberts1]

Has there been any thought to reducing the time limit for a test from 10 days per block to something smaller, or perhaps placing a limit on the total time for a test. i don't think there are any machines out there that can't complete a single boock in even over 2 hours ofr the slowest machines. It is a shame that we can't look at the upper bound and get a good idea of how far we've gotten.
On a seperate note, is there any way to stop a test and send the results in so that other people can start where you left off? If s oI'd like ot know how to do it, if not maybe that'd be a good idea. :|ot|:

[eatmadustch]

My P2 266MHz takes 4 hours :p But I've taken it off sob, it just takes too long to complete a complete test and put it on dnetc.

yes, you can move a test to another computer. If it's windows, you have to move the registry keys (I don't know which ones, do a search in this forum) and the zXXXXXXX file you were working on. If you're useing linux, just move the cache and zXXXXXXX file!

[looxix]

Don't forget there are, for various reasons, lots of machine not 24/24 connected.

Looxix

[Jwb52z]

Keroberts1, you are somewhat mistaken about the 10 day rule. It simply says that the server must hear from your client running a test once very 10 days at least or it will reassign the test to someone else. You don't just complete one block every 10 days.

[Marco_N]

On the other hand, when I read the description for "current test window" (http://www.seventeenorbust.com/stats/rangeStatsEx.mhtml) correctly, for quite a few k the n lower bound is in the 400,000 bracket and the current served tests are around 3,500,000.

Are there really clients still working on n's around 400,000? When were these tests served?

The question here is: Should one test be allowed to drag on for months or even years? Should that test be "The One n To Rule Them All", a lot of clients are wasting a lot of CPU power.

When a block is submitted every 9 days, a 150 block test will take 150 x 9 = 1,350 days, or 3.7 years. Don't get me wrong, I appreciate people putting their PII/233s to good use, but I don't think the current mechanism is very appropriate.

A possible solution may be to reserve the lowest k (4847) for slow machines under the current limits (10 days/block), and the other k's under a new 'per test' limit (for instance: max 20 days/test).

- Marco

PS - in my postings, "test"= k/n pair and block = part of a test.

PPS - the last time I looked, there were only 22 tests left <3,000,000 so why not manually release these 22 tests that are pending for over a month and get this discussion over with :rolleyes:

[Jwb52z]

Surely you are not suggesting that the people in charge of the site force people to complete a test in such a limited amount of time? I understand not wanting a test to take over 3 years, but neither do I think we should be forced to complete a test within 20 days. Some of us run quite a number of different DC projects. If I were limited to that short a time with this project per test for it, I could not finish it with all the other projects I do.

[Samidoost]

>>>
A possible solution may be to reserve the lowest k (4847) for slow machines ....
<<<

:shocked:

That lowest k is the most important, thus needs the fastest machines.

Just a note
Payam

[Moo_the_cow]

Why is the lowest k (4847) the most important?

[Samidoost]

Because, it is *the* first different k, among all the odd numbers.

See http://sierpinski.insider.com/4847_history for a brief history of such k's.

- Payam

PS. Considering the dual (http://sierpinski.insider.com/dual) Proths, the most important is k=28433.

[smh]

Are there really clients still working on n's around 400,000? When were these tests served?

Yes there are. But it's just a double checking process. SoB didn't start testing all K's with N=1. This has been done by others years ago. It's just to confirm that they didn't (accidentely) miss a small prime.

If you change your username to 'secret' then the server will start assigning you these low tests. Chances to find a prime are very small of course, but the tests are short. It took only a couple of computers to test all K's upto N=300.000 in a day or 2 (3?).

I do agree with you that the 10 day per block limmit is to long. This is discussed a couple of times on the forum already.

It might be an idea to check if a test is among the say 100 smallest available, and if so, do not assign them to regular users, but only to a few users which are enabled to get these tests?

[Mystwalker]

A possible solution may be to reserve the lowest k (4847) for slow machines

The k value is irrelevant for the test length. The biggest k is 14.x times bigger than the lowest - this means that the n value has to be only 4 bigger to compensate.

[Marco_N]

Originally posted by Jwb52z
Surely you are not suggesting that the people in charge of the site force people to complete a test in such a limited amount of time? I understand not wanting a test to take over 3 years, but neither do I think we should be forced to complete a test within 20 days. Some of us run quite a number of different DC projects. If I were limited to that short a time with this project per test for it, I could not finish it with all the other projects I do.

That's exactly what I'm suggesting..... but only for all but one k, reserved for this class of participants. Let's nominate one. Since Samidoost (and perhaps rightly so) objected to 4847, why not take 67607.
As long as you agree to work on 67607 only, everything can remain the same (1 block/10 days, only an implied limit on test duration). You will have to take the (slim) chance that someone has been assigned a prime-n and may be taking a (very) long time to complete it. For all other k, more strict rules may be applied. But you'll know you only risk "wasting" a far more limited amount of resources.

Perhaps the easiest way is to make a selection in Preferences, so you won't have to change any client settings - the client won't even know anything has changed except it only gets k=67607 all the time.

- Marco

[Jwb52z]

I hope to God that that never is done by the creators, Marco. Anyone who only does ONE PART of a test every 10 days probably shouldn't be in the search anyway.

[Keroberts1]

Iagree whole heartedly that if it takes someone more than a month ot finish a test, even if they are spreading their resources out, then they shouldn't be taking tests.

[vaughan]

I disagree. My slowest computer is a P166MMX with 128MB RAM and it completes a block in about 5 hours. The problem is the number of blocks in a test is now much larger than when I started in this project more than a year ago. However, my dinosaur is slowly crunching away 24/7 and touches base with the server several times a day so I don't see why the project should exclude "old faithful".

[Marco_N]

On the one hand, I agree every system is (or should be) welcome and appreciated. On the other hand efficient use of available SoB resources should be taken into consideration.

Suppose Old Faithful gets a "prime" n assigned (hurray!) of 175 blocks. It will work for 37 days to confirm this. An equivalent of 3 days of full SoB power will be spent on this k during this time (*). At 1 block / 10 hours this value doubles. At 1 block / 9 days, this is cause for concern.

IMHO it wouldn't be so bad if all 'slow' machines were making themselves useful on cracking 67607 while all other machines were taking care of the remaining k.

- Marco


(*)
Suppose each day 100 P90 years is available and the assigned n's are evenly distributed over the remaining k's (at the time of writing, the distribution is skewed: 55459 - 369 pending; 67607 - 95 pending). The available resources "per k, per day" would therefore be 100/12 = 8.33 P90 years. During a period of 37 days this would mean 37x8.33 = 308 P90 years are used 'in vain'. This is equivalent to 308/100= 3 days of the entire SoB resource pool. Yes, this very simplistic.

[smh]

Suppose Old Faithful gets a "prime" n assigned (hurray!) of 175 blocks. It will work for 37 days to confirm this. An equivalent of 3 days of full SoB power will be spent on this k during this time (*). At 1 block / 10 hours this value doubles. At 1 block / 9 days, this is cause for concern.


Block times are about constant IIRC.

[Marco_N]

What I meant to say, there may be even older Faithfuls out there that take longer to complete a block or newer Faithfuls that aren't on 24/7.
I have unfortunately no insight in the tests due, so maybe we're having a discussion about a non-issue! Perhaps this can be another stats item?

Top 10 of "tests in progress"?
1. Slowlybutsurely k= n= 5m 4d 7:30
2. Littlefaster75Mhz k= n= 4m 1d 0:00

etc.

Or will this level of exposure encourage ppl to be slow on purpose?

:D