| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: do more by doing less? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 16
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
suggestion/possiblity ??
if we can see that valid units are returned (feature currently part of boinc only asfaik) then perhaps we should dump ( voluntarily) incomplete jobs to start on the next job, ie partly pointless if results already validated?? could there be 1 extra w/u sent if the 1st 3 dont addup as valid?? could points be awarded for dumpered workunits with time credited (now there is a longshot) .. at a sacrifice of some points it might be worthwhile for some to abort/dump w/u's once we see valid resutls and thus accelerate the finish time of the project? wot do others think? i would rather miss a few points but get started on work pending rather than duplicate already validated results...i must say i am getting to like the extra info i can see in boinc plus a little bit of configurability is nice... boinc an aquired taste perhaps as one of our admin/adviser said recently ( or was it a reg member?) |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
or even better how about having things setup so that once 3 results are confirmed as valid there is an option to dump "extra" w/u and credit the user regardless and give them a new w/u ( of course only for those connected 24/7) and only as an option if user wants it, could be a way to grant points as a sort of bonus extra then users can get started immediately on a new unit.. maybe setup a feature in boinc ?? boinc seems to be able to "chat" to server hq easy enuf, is it possible to set an option to allow this? perhaps it could save up to 25% of work time in theory if all did it, of course that would have to be for zero time waste, more likely to be way less than that, or perhaps another option (?) would be to hold back the "4th" w/u and send only if needed for validation?
|
||
|
|
David Autumns
Ace Cruncher UK Joined: Nov 16, 2004 Post Count: 11062 Status: Offline Project Badges: 
|
I like this idea
----------------------------------------perhaps another option (?) would be to hold back the "4th" w/u and send only if needed for validation BOINC would appear to already do this if a work unit is taking more than the usual length of time to get a quorum. I'm still waiting for a 135/0 work unit that I and another cruncher returned on the 14th. Yesterday BOINC sent out a 6th WU to another cruncher probably in the hope of getting a reply Dave  [Edit 1 times, last edit by David Autumns at Apr 22, 2006 9:14:46 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
thanx for answer, i will push you a little bit harder...
wot is your opinion of dumping of workunits once validation has been achieved?? is there a point or not? i personally am not that fussed about points, fun as they are, i would rather see the projects completed faster and there are probably ways to do this.. in terms of efficiency of throughput.. i note the quorum of 3 seems to be the go instead of previous quorum of 4, many tiems i see 3 w/u returned by various users then a valid stamp being given but some other user is still struggling with this already validated w/u.... ps i also noted today a group of 4 returned units all pending validation, am interested (slightly) in the process, is it a simple - they agree -therefore they are ok - principle or is there more to it? boinc seems to give eyes into progress of other w/u so if a process could be automated then there might be a win/win situation possible, ie if user online and the boinc client chats to the server and finds the w/u already valid then another w/u can start ealrier... cheers.. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
well here ya go, this is wot i did today, i looked at my old linux disk - ubuntu ( cos i am changing to suse on a different disk but same computer) . i pluggen the old disk back in for a look, i saw 2 work units in progress ( around 3 or 4 days old) and 1 pending i checked in boinc re the status of them, as i saw them all having been validated now i decided to dump them in favour of new work ( which will be done by same computer but diffrent linux install ie suse, ) even though they were well into job i decided to lode some points on those not defunct w/u and launch into new work,, less points but more work IMHO (?) cheers... any comments??? is this good idea - dumb idea? do i have a wrong persceptive??? cheers
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Sadly, I believe we are dealing with a lost cause on humanitys part. Research projects like this have a "finite" amount of people willing to dedicate resources to help this effort.
Look at MySpace...millions of people in selfish interests on there. With just half them resources half the world problems would be solved overnight. Everyone seems to know someone on myspace nearly. Projects like this will benefit most from technological gains in my opininon. AMD is suppose to release the first CPU core able to process 4 threads at the same time or a 4 core processor. Talks of 6 cores are in the pipeline as well. This would be early to mid 2007 In the end there are only enough humble people willing to dedicate to worth while projects such as this...the rest we depend on technology to solve it. So technology allows us to do more with less in short. Opposition of views welcome..just no flaming Have a good day |
|||
|
||||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Sadly, I believe we are dealing with a lost cause on humanitys part. Research projects like this have a "finite" amount of people willing to dedicate resources to help this effort....... Opposition of views welcome..just no flaming Have a good day i would very much like to disagree with you, UNFORTUNATLEY i cant :( i amsurrounded by lots of clever people who work in the health casre industry - can you believe it!, so far of the 50 or so i have mentioned this wcg etc projects to, NOT ONE has joinded up asfaik... the main excuse is oh my puter is too important to waste timne on it, or i am too worried that it is a cia plot or other company tkaing over your computer, drive me nuts.. then the usual thing that people would rather be wrong in a group than to stand ontheir own and fight for a cause... my son has joined us and my father in law ( is on a separate project but i plan on swapping him to here sooner or later) .... on a more positive note i think if we can get some real ersults happening then the masses might come and help , but as usual research takes years to filter thru to the real world so we will struggle on regardless, i have to say that 100 years of comptuer time everyday or wotever it is is still an amazing thing and i am sure it will be helpful to humanity.. in the meantime i am looking at ways i can make my own little contribution just that little bit more efficient cheers |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Unless I got it all wrong with regard to how Rosetta works (and I'm pretty sure I don't), HPF shouldn't need to use redundancy or quorum for any reason other than verifying credit claims.
----------------------------------------I really don't understand why WCG operates with such high redundancy and quorum numbers, whereas it could triple-quadruple (3x - 4x) the overall science done. [Edit 1 times, last edit by Former Member at Apr 26, 2006 1:19:16 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
 Maybe, maybe not. I can imagine a CPU that makes floating point errors in almost every conformation, returning 100-500 structural predictions in a HPF result that are subtly off. One way of predicting the likelihood of a particular structure is to examine homologous proteins in several dozen different species in which the analogous protein (the homolog) differs at several points. This might produce many radically different predictions across species, but if each homolog produces n structural predictions found in each species, then it is very likely that the correct prediction will be one of those n structures. Including invalid structural predictions could cause problems if you tried to use homologous structural prediction. There are other possible methods discussed over at Baker lab that are also sensitive to errors. I do not think that an occasional floating point error caused by a cosmic ray would really matter to this project, but I suspect that some people have very flaky computers. Any way, redundancy is specified by project scientists after getting the Lecture on our historical error rates. As we have been gaining history, the redundancy requirements have been dropping. Lawrence |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The error rate is real. Small, maybe - but real. To run through the possibilities: processor error, memory errors, disk corruption and upload errors.
A few people who have been overclocking beyond sensible limits have experienced a 100% failure rate. The majority of people, however, will never see an error. For this kind of science, it isn't good enough to say "99.5% of our results are accurate". Triple checking makes the results far more reliable. |
||
|
|
|