Uploading Result

[LinearB]

After 82 hours of solid crunching managed to complete my first unit. Unfortunately I'm on dialup so it couldn't upload the result :(

A few hours pass.......

Having gone back online a suitable unit was downloaded and the crunching started again in earnest, but it doesn't appear to have uploaded the completed unit (nothing in the stats, and not mentioned in the logs).

Suggestions please :help:

[jjjjL]

you should still have the zXXXXXXX file if it didn't submit it. the client gives up trying to submit a block sometimes if it times out too much. it's so that that it can continue work even if a block is denied but i suppose the client should differentiate between not being to open a connection and not being able to work with the server.

email it to me and i'll put it in the db for you.

-Louie

[smh]

The same thing happened to me last weekend. A machine completed a proth test but wasn't able to contact the server. I think the client tries a couple of times with 5 minutes interval (300 seconds and writing a line to the log file every second!!!)

Luckily i had a reg file for that exponent. Stoped the new test, exit the client, run the reg and start the client again. It continued at 99.9% and had to do the last 10 minutes again before it reported the result. After that it got a new test which i abandoned since the other test was already a couple of hours underway before i found out. This means another exponent i have reserved for two weeks but won't complete.

I know it isn't a big deal, it's only two weeks, but it would be nice if the client had an option to quit SoB.

How many exponents till we reach 3M? about 2250 i guess?

You think you'll be able to make a list available with the exponents that stil need testing? (even beter, also a list with tested exponents and their (masked) residue

[LinearB]

Thanx Louie have sent email here:-

Hope this is right ???

lhelm@umich.edu

[jjjjL]

smh - yeah, that works well. i didn't feel like explaining that all.
regedit isn't for all users.

1850 tests below 3 million.

the system completes about 235 tests every week so we're on track for finishing 27653 this year still. and since we're always testing around 200 blocks at a time, we should start in on k=33661 even sooner.

there are only 7 below 2 mill.

they always seem to get assigned to the wrong folks and never get done.


things are going well. back to working on the client.. later.

-Louie

[smh]

Originally posted by jjjjL
smh - yeah, that works well. i didn't feel like explaining that all.
regedit isn't for all users.

Nope definately not for all users. That might be a good reason to use ini files instead of the registry (although reg files are much easier if you know how it works ). Not everybody has rights to add a key to the registry when installing the program.

I just looked in the log file, the client tries to report the result 5 times with 5 minutes interval. (do we really need a line in the log every second?) but after that never tries again. I think this is a serious problem and should be fixed in a future client. It should be reported when it gets the next test.


Originally posted by jjjjL
1850 tests below 3 million.


Guess i miscalculation by me, i thought we would need about 8000 test to reach 3M, guess this will be around 7600 then.


Originally posted by jjjjL
the system completes about 235 tests every week so we're on track for finishing 27653 this year still. and since we're always testing around 200 blocks at a time, we should start in on k=33661 even sooner.


Don't forget things will slow down a little when the numbers get larger. Not sure if this compensates the increment of users. If one of the remaining numbers is prime it will be the 4th or 5th largest known. When we go to the next K test will take only a few hours. I think this is a time to attract some new users. Many people don't want to wait days for a test is done (this is why i'm not running the GIMPS client anymore, at least not for LL testing)



Originally posted by jjjjL
there are only 7 below 2 mill.

they always seem to get assigned to the wrong folks and never get done.


I have one of them. Don't worry, it should be completed in 9 hours. Hmm, now i think of it, probably a few hours longer coz i'm thinking about transfering it to another computer coz i'm not able to contact the server at the moment (problem on my side)

Sander

[alpha]

Originally posted by smh

<snip>
I think the client tries a couple of times with 5 minutes interval (300 seconds and writing a line to the log file every second!!!)
<snip>

I mentioned this in http://www.free-dc.org/forum/showthread.php3?s=&threadid=1584
perhaps it was missed :confused:

[jjjjL]

i am very much aware of the "line for every second" *feature* ;).

i know it's dumb. it was like the first thing i fixed after i released v0.9.7. so the new version i'm playing with definately doesn't do that.

ok, about the retries... yeah, those values are a bad. they made more sense when the project started but no longer. you should all change your report retries to 0 the wait time to something more reasonable like 3600 (1 hour). 0 retries = infinite.... undocumented *feature*... there will be a note in the next version's config <adds it to code right now>.

i just got in-client signup working too. :) there were some wicked problems with it for awhile. blocking also does less insane things to the stats graphs too. :)

now would be the time to suggest block sizes for the next client as I am planning to build it into the client. i'm thinking ~1 hour on my computer (1.33GHz Athlon). This would put most intel boxes at around 2 hours.... in all, everyone will be well under a day.

Have to balance -- smaller blocks = some more bandwidth , more pauses if the computer isn't online, but also better online stats :)

i've thought about having it be user defined, which wouldn't be too hard, but i think that deminishes the point of comparing block times a bit.

oh well... think about it people. i'll watch for replys. don't forget to update your retry values in your client. next versions defaults will be 0/0/3600 but i don't plan to make it update current user values by default.

-Louie

[Firebirth]

i've thought about having it be user defined, which wouldn't be too hard, but i think that deminishes the point of comparing block times a bit.

I like the point that you get your "own" number! - So if I were to decide, I would make it two options for the size of blocks! One where you get your own "perhaps-prime" number! And one where you get the smaller, and thus more efficient blocks (I guess it is quite fine with standard size of these blocks!).

[smh]

Originally posted by alpha


I mentioned this in http://www.free-dc.org/forum/showthread.php3?s=&threadid=1584
perhaps it was missed :confused:

I missed it, but you're cause is a bit different then mine. Louis explains why your blocks were denied.

Mine says, can't connect. I can explain why i couldn't connect, but not why the block didn't report. (had the same problem this morning, but now i was watching the last minute of the test). When our i-net proxy was up and running again i had to rerun the last 10 minutes of the test before the result was reported.

Any way, i changed the settings to 0,0,300 (ya, 300, don't want to wast to much time in case the connection comes up again)

[smh]

now would be the time to suggest block sizes for the next client as I am planning to build it into the client. i'm thinking ~1 hour on my computer (1.33GHz Athlon). This would put most intel boxes at around 2 hours.... in all, everyone will be well under a day.

Actually i don't care much about blocks and stats etc.. (you have to do a whole test anyway!). But it would be good if it's used to expire exponents. I think you should bring the expiration time back to a couple of days (a week max) since the last check, in if you make 1 hour (1,33G Atholon) blocks. A slow PII can do such a block within a day even if it's not on 24/7 (like my PIII 400).

But please don't forget that not all clients are online all the time so don't let the client wait until it can report but continuing crunching in the mean time!!

[priwo]

I am online only once a day (modem-connection) and can do the actual blocks in about 30 hours on my P4. I like the current block size.

priwo

[smh]

Originally posted by priwo
I am online only once a day (modem-connection) and can do the actual blocks in about 30 hours on my P4. I like the current block size.


My notebook takes over 100 hours for current tests and is only on for about 40 hours a week. If i'm not able to let the server know that i'm still running a test i'm not able to complete an exponent withing the two week resevation period.

An other question for Louis.

How do you calculate the size of one block? Is it a percentage of the whole test, or a number of iterations, or.....?

If it's a percentage then it means that blocks are getting longer and longer as the numbers we test are getting bigger, and suddenly, when we change to the next K the blocks will be very short in the beginning (10 mins on a fast P4?)

If it's a number of iterations there will be more blocks in a current test compared to when we test much smaller numbers, but each iteration will also be faster if the program uses a smaller FFT.

The latter seems more honest then the first, but a small number to test will give more blocks in a timeframe.

Or is there an other method?

Sander

[jjjjL]

blocks will be constant cEMs.

right now my test client is using 100kcEMs / block.

this will stay about constant across different k/n values.

cEMs aren't a perfectly constant measure, but they are close.

and it will actually work exactly the opposite of how you describe... lower n values will have slightly longer blocks (by a small amount i think.. i'll test and see).


also, good news, I have essentially finished 0.9.8, testing aside.

teams are done, mem leak fixed, blocking done, logging fixed, signup in client finished, progress is saved on exit and stop, retry code improved, installer bug swashed, and tons of other cool and new features.

and just to tease... ;)

-Louie

http://www.seventeenorbust.com/images/screenshot.png

[smh]

right now my test client is using 100kcEMs / block

Hmmm..., Guess my understanding of blocks is not really correct. With 82K cEMs / sec your pc is doing 1 block every second or so?

Did you made a typo or should i read the website again?

Sander

[jjjjL]

100M cEMs, not k cEMs.

so it takes me around 20 minutes.

-L

[Halon50]

To help reduce a forseeable server load when you implement blocks, would you consider a slightly larger block size? Say, 500McEMs? This would take, I'd say, around 4 hours to complete one block on my P3-450, which is certainly within my range of "acceptability," while machines like my XP2200+ would go from 12 minutes (on your scale) to around an hour.

With several hundred machines spewing blocks every hour, wouldn't this small blocksize cause some bandwidth spikes on your end?