51 hour runtime?
18 Mar 2014 5:57:09 UTC
Topic 197458
(moderation:
Oooh - hi. Sorry to trouble you but just downloaded four 51 hour tasks. Never had any that long before. :/ Slightly concerned because my computer tends to go a little bit bleuch when running more than one einstein at a time... and once a cosmology unit is chucked into the mix (am doing one at the moment which I think will be with me forever)... bleuch no longer covers the mind-boggling depravity my computer is capable of.
Will go away and do some maths to work out if there are anough hours before the 1st April, given my computer is not on all the time, but if I do run out of time - do I continue crunching late units or do I abort them? Actually, I could have made this a much shorter question by just asking that couldn't I? Sorry. :) Best wishes and thank you for your time.
Please wait here. Further instructions could pile up at any time. Thank you.
51 hour runtime?
)
No problem with the length of the post, it's usually better to give a bit to much info than to little. =)
The new tasks you downloaded are "Gamma-ray pulsar search #3" and yes they do take more time to process than "Gravitational Wave S6" tasks.
Remember that the "Time to completion" is an estimate and might be wrong, to get a better feel for the time it's going to take you could do you own calculation once the tasks reaches 10% or so done.
If you don't want to run Gamma-Ray tasks in the future then go to your Einstein@Home preferences and opt out.
As for tasks that might be late, if you can finish them and report them before they get sent out to somebody else or at least before that person returns it you will get credit for the work. I would only consider aborting tasks if they have only just started running at close to the deadline or if the third person have already returned their copy.
Thank you! Will give them a
)
Thank you! Will give them a go and soldier on! Who knows... might add that extra bit of a challenge :) can't compete against the big fish re points so can compete against myself and the clock instead! Best wishes Anniet
Please wait here. Further instructions could pile up at any time. Thank you.
RE: Thank you! Will give
)
Just remember most of the "big fish" started with one machine in the beginning too!
RE: Thank you! Will give
)
Yep, that's how it starts, you get bitten by the bug and strive to improve, then before you know it you are hooked. The focus of the challenge then changes to...
Where am I going to put host 2,3,4..57, cover the cost of electric and pay for the latest greatest graphics cards and other hardware?!!
As the saying goes - from a tiny seed, a mighty oak grows :-)
You have to start somewhere and rest assured that the minnow's at Einstein have an equal chance of discovering something new, as the great white fishes do ;-)
Gavin.
Your encouragement is so
)
Your encouragement is so sweet! Thank you! I can see it now... multiple shiny computers, simultaneously indulging in mind boggling depravity. :) Can't wait!
While I'm here, the first task is at 57% after 11 hours! Which is very encouraging! Of course it does suggest that a computer somewhere at einstein looked at my system... and was then unecessarily rude about it :) Thank you all for not being so! :)
Best wishes and happy crunching!
Anniet
Please wait here. Further instructions could pile up at any time. Thank you.
The 51 hours was an initial
)
The 51 hours was an initial estimate by the BOINC client software on your rig, for the given work unit type, which will adjust downwards based upon your rig's recent actual performance.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: Your encouragement is
)
Sort kinda, maybe, probably not so much as you think. As Mike said the Boinc software we all installed does that for each project that you sign up for, telling the projects servers what hardware is in your machine and what kind of work units it needs based on that hardware. So if you have an Nvidia graphics card you would not get AMD graphics work units, you would get Nvidia work units. This way your system gets what it needs and can handle and not stuff it can't handle. What if you don't have a graphics cards that can crunch though...it knows that too and won't send you any of those kinds of units at all, and will only send you cpu units instead. Every computer has a cpu but some are faster then others, Boinc runs a short basic test of your computers capabilities and then gives a rough ballpark estimate of how long each unit will take to finish. As you can probably already see that test needs work as it is usually wrong in its ballpark guesstimate. BUT the key is it does sorta kinda work and it's better to send you too little work then too much work. You can always get more work, it's kind of hard to take back what has already been sent out on an individual pc by pc basis.
Mike Hewson wrote:The 51
)
Uhm, the 51 hours initial estimate is only used by BOINC because of the generic amount of fpops that the Einstein project sends out with these tasks. BOINC will then take that rsc_fpops_est (resource floating operations per second estimate) value, divide it by the CPU's floating point benchmark value, and then you have an estimate of the run time.
Without the rsc_fpops_est, BOINC won't be able to give an estimate of the run time.
Just pointing out that it's not the BOINC client on its own that does the estimating, which is something you seem to indicate in your post. Also pointing out that the rsc_fpops_est value is the same for all computers it is sent to, be it old P3s, or the latest Haswell, be it those with and those without GPUs. The value should be derived with the help of a test computer, that just runs tasks. Then taking the run time of those tasks, averaging the run time over a couple hundred of them, you can recalculate the average floating operations that that took, and use that as a base value for tasks.
Each computer running BOINC will then adjust that runtime through running several of those tasks in a row. By using the DCF value in the client (Duration Correction Factor), can the BOINC client adjust the run time estimate up- and downwards. Not all projects use the DCF value anymore, though. They calculate the difference on the server based on the actual run time of results that were returned, and send that value along with any next scheduler contact. Einstein is a project that doesn't use DCF anymore.
RE: Einstein is a project
)
Er, really? That's news to me.
I know Einstein has a long-term objective of moving to the newer server code which performs the adjustments centrally (and much more slowly), but it hasn't got there yet. It's still the same old, same old here: which makes it difficult (nay, impossible) to accurately adjust for all of the different applications in use here at the same time.
I checked my client_state.xml
)
I checked my client_state.xml file prior to posting that, it has the same entry for Einstein that Seti has, which I thought meant that DCF was not used.