There's a difference between the client discarding work because of a user 
finger-fumble, and the server discarding it because of an as-yet unknown RPC 
problem.

If the client discards work, it is either 'resent lost results', or allowed to 
time out with no response - depending on server configuration. Nothing is lost.

But if the *server* marks the work as 'abandoned', but the client retains it 
and continues to work on it, then electricity is wasted.



>________________________________
> From: Jon Sonntag <[email protected]>
>To: Richard Haselgrove <[email protected]> 
>Cc: "[email protected]" <[email protected]> 
>Sent: Sunday, 4 May 2014, 0:15
>Subject: Re: [boinc_dev] Deadlines
> 
>
>*I didn't have any of the reliable* config settings
>for accelerating retries set, but in case it was doing it automatically or
>becomes of some other settings, I added *reliable_reduced_delay_bound with
>a value of 1.0 so if that was the case, it should now be ignored.
>
>As far as abandoned WUs go, a large number of the Collatz users have
>app_config files and/or app_info files and if there is an error, the host
>abandons all the WUs at startup.  I don't think the server ever gets
>notified when that happens.  The WUs just disappear and the files are
>deleted when BOINC starts.  Maybe it would be better if it kept them around
>and aborted them or, better yet, put them on hold somehow as the user
>generally figures out the error and fixes it.
>
>Jon Sonntag
>
>
>On Sat, May 3, 2014 at 3:47 AM, Richard Haselgrove <
>[email protected]> wrote:
>
>> The workunit in question,
>> http://boinc.thesonntags.com/collatz/workunit.php?wuid=5741706, has two
>> results: the first had an initial deadline of 8 May (normal), and it's only
>> the retry which has the reduced 2-day deadline. I imagine David's reference
>> to http://boinc.berkeley.edu/trac/wiki/ProjectOptions#Acceleratingretriesis 
>> the likely explanation.
>>
>> But it raises a secondary question. Why does the first user's computer
>> have 200 tasks marked as 'Abandoned'?
>>
>> http://boinc.thesonntags.com/collatz/results.php?hostid=115599&state=6
>>
>>
>> This seems to be a relatively recent phenomenon (timescale 1 year / 18
>> months). It can happen by deliberate user action (detach from project), but
>> in every case I've investigated following a user report on a message board,
>> the user has been adamant that they didn't detach or take any action which
>> should plausibly lead to tasks being abandoned: indeed, they report that
>> there is no sign on their computer that anything is wrong, and the tasks
>> are still shown in BOINC Manager and are being computed and reported as
>> normal. It's only when/if they visit their project account page - perhaps
>> because they notice a reduced rate of credit being awarded - that they find
>> their time and electricity is being used to no purpose.
>>
>> mark_results_over() in handle_request.cpp is supposed to be called when
>> there is 'evidence' that the host has been detached/reattached, the
>> statefile has been copied from another machine (corrupting rpc_seqno), or
>> some major event like that. But it seems to happen in other cases too -
>> most recently http://www.gpugrid.net/forum_thread.php?id=3740#36629.
>>
>> There seems to be some correlation in client logs with failed RPC
>> attempts, perhaps reinforcing the rpc_seqno theory, but it really needs
>> somebody with access to a server log to look into this. It greatly annoys
>> the volunteers when they find a substantial volume of work (as in this
>> case) has been thrown down the drain.
>>
>>
>>
>> >________________________________
>> > From: David Anderson <[email protected]>
>> >To: [email protected]
>> >Sent: Saturday, 3 May 2014, 6:01
>> >Subject: Re: [boinc_dev] Deadlines
>> >
>> >
>> >The scheduler has an optional mechanism that reduces the latency bound
>> >of results that
>> >1) are retries
>> >2) are being sent to a "reliable" host
>> >See:
>> >http://boinc.berkeley.edu/trac/wiki/ProjectOptions#Acceleratingretries
>> >
>> >Check your config.xml to see if you're using this.
>> >
>> >-- David
>> >
>> >On 02-May-2014 9:41 PM, Jon Sonntag wrote:
>> >> Anyone have any idea why a result would have a deadline of 2 days when
>> the
>> >> work generator has the delay bound set to 7 * 86400 or 7 days?
>> >>
>> >> Created2 May 2014, 19:57:23 UTCSent2 May 2014, 21:07:30 UTCReport
>> deadline4
>> >> May 2014, 3:36:58 UTCReceived3 May 2014, 2:30:23 UTCLast time modified2
>> May
>> >> 2014, 21:30:26 UTC
>> >> The workunit has the following info according to the database:
>> >>
>> >> mysql> select FROM_UNIXTIME(create_time),FROM_UNIXTIME(delay_bound) from
>> >> workunit where id=5741706;
>> >> +----------------------------+----------------------------+
>> >> | FROM_UNIXTIME(create_time) | FROM_UNIXTIME(delay_bound) |
>> >> +----------------------------+----------------------------+
>> >> | 2014-04-30 18:40:23        | 1970-01-14 18:00:00        |
>> >> +----------------------------+----------------------------+
>> >>
>> >> Even if the deadline was related to the time the workunit record was
>> >> created, it would still be May 6 and not May 4.  Anyone have a theory?
>> It
>> >> seems to be happening randomly  For large WUs, it causes the client to
>> >> panic and go into high priority mode and the end users are not happy
>> when
>> >> that happens.
>> >> .
>> >> Jon Sonntag
>> >> _______________________________________________
>> >> 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.

Reply via email to