I guess it's somehow a semantic confusion. WorkUnit != RESULT.
*You* create a *Work Unit* (aka WU, aka JOB). If you trust your machines you can instruct BOINC to generate only one *RESULT* per WU that you create. But some projects create 2 or more *RESULT* per WU to compare them (if two or more unrelated machines arrive to the same conclusion then assume the *Result* they arrive it's *valid*). If for some reason (because BOINC is designed to work with unreliable nodes) a *RESULT* it's lost (because it timed out, exited with an error, did not match another *Result* for the same work unit) then the BOINC framework generates more results for this work unit and send them to the clients, until a consensus is reached (via validator) or another work unit limit it's reached (WU_ERROR_TOO_MANY_ERROR_RESULTS, WU_ERROR_TOO_MANY_TOTAL_RESULTS). A *RESULT* it's never re-sended, but another *RESULT* it's created if needed instead. A valid *RESULT* it's called "canonical_result" and it's considered *the work unit result* I hope it helped 2013/6/21 McLeod, John <[email protected]> > It is not really what is going on at the server end from the client end. > We know how it works overall, so we can see how the details mesh. This > would not really be obvious just by watching the client. > > -----Original Message----- > From: Charles Elliott [mailto:[email protected]] > Sent: Friday, June 21, 2013 7:51 AM > To: McLeod, John; 'Chanda Sarkar'; 'Christian Beer' > Cc: [email protected] > Subject: RE: [boinc_dev] Questions around BOINC > > Instead of all these questions, why don't you just attach a computer to > Boinc's SETI@Home project, or another -- there are tens of them -- and > watch > how it works. Your questions seem to fundamental and trivial to those who > have been using Boinc for years. > > Charles Elliott > > > -----Original Message----- > > From: boinc_dev [mailto:[email protected]] On Behalf > > Of McLeod, John > > Sent: Thursday, June 20, 2013 11:53 AM > > To: Chanda Sarkar; Christian Beer > > Cc: [email protected] > > Subject: Re: [boinc_dev] Questions around BOINC > > > > Each work unit will be made into one or more identical tasks depending > > on the initial replication setting. Each of these tasks will be sent > > to a different computer. When the task on an individual computer is > > completed, the result of the computation will be uploaded. If more > > than one task was sent, the results will be compared in the validator. > > > > Some applications can do a quick reverse computation to see if a result > > is valid or not, others require two results of the same computation to > > compare to each other. Unfortunately, individual computers cannot be > > trusted as they may overheat, be overclocked, have a random computation > > error, or (grumble) be maliciously modified so that credit accumulates > > faster. Hence the need for validation of computational results. > > > > SETI Classic had a handful of clients that were modified such that they > > would return pre-calculated results as quickly as the system would > > allow so that they would gain credits faster - thus potentially harming > > the science being done. This handful of clients returned a large > > number of invalid results. The same thing could happen in BOINC if > > there is no validation. > > > > -----Original Message----- > > From: boinc_dev [mailto:[email protected]] On Behalf > > Of Chanda Sarkar > > Sent: Thursday, June 20, 2013 7:42 AM > > To: Christian Beer > > Cc: [email protected] > > Subject: Re: [boinc_dev] Questions around BOINC > > > > Hi, > > > > Thanks for the information! But the URL that I read - > > http://boinc.berkeley.edu/trac/wiki/JobIn , says, in *JOB *section* - > > '*One > > or more *results*, each of which describes an instance of a > > computation, > > either unstarted, in progress, or completed. *The BOINC client software > > refers to results as "tasks"**'. *So, one work unit(job) may be > > associated > > with one or more results? More over what is the exact meaning of result > > file is it a single output file or a collection of output files that > > refers > > to single work unit. > > > > One more question that comes into my mind comes after reading - > > http://www.irelandboinc.com/how-boinc-works. > > Under Credit section - 'Each work unit may be sent to several > > computers' > > and 'When at least two results have been returned, the server compares > > them. If the results agree, then users are granted credits.'. > > According to this statement, same work unit(job) are > > sent to > > different machines for computation. So, does that mean same work > > unit(or > > job) may span multiple machines?.. which would contradict to the > > earlier > > mail. > > > > Do correct me if I am wrong > > > > Thanks, > > Chanda > > > > > > > > On Thu, Jun 20, 2013 at 2:36 PM, Christian Beer <[email protected]> > > wrote: > > > > > Hi, > > > > > > I'm not Rom but I hope I can help you too. > > > > > > 1. That scenario is already the principal design behind BOINC. Once a > > > task is associated with a host (gets send) this will not change. So > > in > > > your example: 't1' will only be processed on 'M1', if for some reason > > > this is not succesfull the server will create 't3' and send this to > > 'M3'. > > > > > > 2. What you see in your Screenshot is exact the behaviour I described > > > above. 't1' was created when the workunit (job) was created (because > > of > > > initial_replication=1) it got send to Computer 1619876 ('M1') on June > > 13 > > > later (on June 17) this computer reported that it has detached from > > the > > > project and discarded all remaining tasks. So the server created > > another > > > task ('t2') and it got send to 'M2' (1518828). This is also a normal > > > BOINC behaviour. If the first task is successfull there will be no > > > second task because the workunit is considered complete if one task > > is > > > succesfull (minimum_quorum=1). > > > > > > Regards > > > Christian > > > > > > Am 20.06.2013 10:53, schrieb Chanda Sarkar: > > > > Hi Rom, > > > > > > > > I have two questions around BOINC - > > > > > > > > 1. Is there any chance that single task within a work unit is > > > > processed only once on a single machine and no other machine. > > > > For example : > > > > Assume there is a work unit 'A' that has two tasks 't1' and 't2'. > > 't1' is > > > > send to machine 'M1' for processing and 't2' is send to machine > > 'M2' for > > > > processing. Is it a possibility that 't1' which is processed on > > machine > > > > 'M1' will never be processed for any other machine ? > > > > > > > > > > > > 2. Is there any possibility that task within a work unit may > > be > > > > generated with unpredictable time span which could be a day or > > more? I am > > > > attaching a snap shot for the same > > > > URL : http://screencast.com/t/HOvebaVpP0N > > > > In the above image it is seen for a work unit > > > > > > 'cryo_bk__chain_I_subrun_000_SAVE_ALL_OUT_IGNORE_THE_REST_86456_479' > > that > > > > has two tasks. One is created on 13th June 2013 and other on 17th > > June > > > > 2013. Is there a possibility that tasks within work units can be > > created > > > > with unpredictable time span? Please correct me if I am wrong. > > > > > > > > > > > > Thanks, > > > > Chanda > > > > _______________________________________________ > > > > boinc_dev mailing list > > > > [email protected] > > > > http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev > > > > To unsubscribe, visit the above URL and > > > > (near bottom of page) enter your email address. > > > > > > > > _______________________________________________ > > boinc_dev mailing list > > [email protected] > > http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev > > To unsubscribe, visit the above URL and > > (near bottom of page) enter your email address. > > _______________________________________________ > > boinc_dev mailing list > > [email protected] > > http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev > > To unsubscribe, visit the above URL and > > (near bottom of page) enter your email address. > > _______________________________________________ > boinc_dev mailing list > [email protected] > http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev > To unsubscribe, visit the above URL and > (near bottom of page) enter your email address. > _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
