"shortfall" is defined for the processor type (e.g. NVIDIA GPU) as a whole,
not per-project.

Is there more to this problem than this?

-- David

On 13-Jun-2012 11:51 AM, Jon Sonntag wrote:
>
> Problem #2: Collatz is set for GPU only and PG is set for CPU only.  Why
> would PG have a shortfall of GPU when it will never get GPU work?
>
> Wed Jun 13 13:17:27 2012 |  | [work_fetch] ------- start work fetch state
> -------
> Wed Jun 13 13:17:27 2012 |  | [work_fetch] target work buffer: 691200.00 +
> 43200.00 sec
> Wed Jun 13 13:17:27 2012 | Collatz Conjecture | [work_fetch] REC 2159.058
> priority 0.000000 (project backoff 49.66)
> Wed Jun 13 13:17:27 2012 | PrimeGrid | [work_fetch] REC 504.749 priority
> -1.097386 (susp via GUI)
> Wed Jun 13 13:17:27 2012 |  | [work_fetch] CPU: shortfall 1468800.00 nidle
> 2.00 saturated 0.00 busy 0.00
> Wed Jun 13 13:17:27 2012 | Collatz Conjecture | [work_fetch] CPU: fetch
> share 0.000 rsc backoff (dt 0.00, inc 0.00) (blocked by prefs)
> Wed Jun 13 13:17:27 2012 | PrimeGrid | [work_fetch] CPU: fetch share 0.000
> rsc backoff (dt 0.00, inc 0.00)
> Wed Jun 13 13:17:27 2012 |  | [work_fetch] NVIDIA: shortfall 734400.00 nidle
> 1.00 saturated 0.00 busy 0.00
> Wed Jun 13 13:17:27 2012 | Collatz Conjecture | [work_fetch] NVIDIA: fetch
> share 0.000 rsc backoff (dt 0.00, inc 0.00) (no apps)
> Wed Jun 13 13:17:27 2012 | PrimeGrid | [work_fetch] NVIDIA: fetch share
> 0.000 rsc backoff (dt 0.00, inc 0.00)
> Wed Jun 13 13:17:27 2012 |  | [work_fetch] ------- end work fetch state
> -------
>
> Problem #3: After detaching from PrimeGrid, it still won't ask for any
> Collatz work even though there is a shortfall:
>
> Wed Jun 13 13:20:27 2012 | Collatz Conjecture | [sched_op] Starting
> scheduler request
> Wed Jun 13 13:20:27 2012 | Collatz Conjecture | [work_fetch] request: CPU
> (0.00 sec, 0.00 inst) NVIDIA (0.00 sec, 0.00 inst)
> Wed Jun 13 13:20:27 2012 | Collatz Conjecture | Sending scheduler request:
> Requested by user.
> Wed Jun 13 13:20:27 2012 | Collatz Conjecture | Not reporting or requesting
> tasks
> Wed Jun 13 13:20:27 2012 | Collatz Conjecture | [sched_op] CPU work request:
> 0.00 seconds; 0.00 devices
> Wed Jun 13 13:20:27 2012 | Collatz Conjecture | [sched_op] NVIDIA work
> request: 0.00 seconds; 0.00 devices
> Wed Jun 13 13:20:28 2012 | Collatz Conjecture | Scheduler request completed
> Wed Jun 13 13:20:28 2012 | Collatz Conjecture | [sched_op] Server version
> 611
> Wed Jun 13 13:20:28 2012 | Collatz Conjecture | Project requested delay of
> 182 seconds
> Wed Jun 13 13:20:28 2012 | Collatz Conjecture | [sched_op] Deferring
> communication for 3 min 1 sec
> Wed Jun 13 13:20:28 2012 | Collatz Conjecture | [sched_op] Reason: requested
> by project
> Wed Jun 13 13:20:28 2012 |  | [work_fetch] Request work fetch: RPC complete
> Wed Jun 13 13:20:33 2012 |  | [work_fetch] work fetch start
> Wed Jun 13 13:20:33 2012 |  | [work_fetch] ------- start work fetch state
> -------
> Wed Jun 13 13:20:33 2012 |  | [work_fetch] target work buffer: 691200.00 +
> 43200.00 sec
> Wed Jun 13 13:20:33 2012 | Collatz Conjecture | [work_fetch] REC 2158.767
> priority 0.000000 (project backoff 176.78)
> Wed Jun 13 13:20:33 2012 |  | [work_fetch] CPU: shortfall 1468800.00 nidle
> 2.00 saturated 0.00 busy 0.00
> Wed Jun 13 13:20:33 2012 | Collatz Conjecture | [work_fetch] CPU: fetch
> share 0.000 rsc backoff (dt 0.00, inc 0.00) (blocked by prefs)
> Wed Jun 13 13:20:33 2012 |  | [work_fetch] NVIDIA: shortfall 734400.00 nidle
> 1.00 saturated 0.00 busy 0.00
> Wed Jun 13 13:20:33 2012 | Collatz Conjecture | [work_fetch] NVIDIA: fetch
> share 0.000 rsc backoff (dt 0.00, inc 0.00) (no apps)
> Wed Jun 13 13:20:33 2012 |  | [work_fetch] ------- end work fetch state
> -------
> Wed Jun 13 13:20:33 2012 |  | [work_fetch] No project chosen for work fetch
>
> Note:  It should be using the anonymous platform and getting OpenCL tasks to
> run on the nVidia GPU.
>
> Issue #4: Since zero debts is deprecated, how does one override the
> calculations and force it to get work?
>
> Jon Sonntag
>
>> -----Original Message-----
>> From: [email protected] [mailto:boinc_dev-
>> [email protected]] On Behalf Of David Anderson
>> Sent: Wednesday, June 13, 2012 1:15 AM
>> To: [email protected]
>> Subject: Re: [boinc_dev] new scheduler config options
>>
>> Robert:
>> To report bugs related to work fetch,
>> set the<work_fetch_debug>  flag in your cc_config.xml file (see
>> http://boinc.berkeley.edu/trac/wiki/ReportBugs)
>> and send me a log segment showing the problem.
>>
>> -- David
>>
>> On 12-Jun-2012 9:22 PM, Robert Miles wrote:
>>> Are there any particular client-end scheduler options for debugging
>>> situations when the client continues running, but stops asking for any
>>> more workunits at all even if it runs out of unfinished workunits?
>>>
>>> My laptop often shows that situation whenever I tell it to maintain
>>> less than a 3-day queue of workunits, even though it is almost always
>>> connected to the internet.
>>>
>>> Robert Miles
>>>
>>>> Date: Thu, 07 Jun 2012 15:38:34 -0700
>>>> From: David Anderson<[email protected]>
>>>> Subject: [boinc_dev] new scheduler config options
>>>> Message-ID:<[email protected]>
>>>> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>>>
>>>> There are two new config.xml options that are useful for debugging
>>>> scheduler crashes:
>>>
>>>> <scheduler_log_buffer>    (lets you set stderr buffering to zero)
>>>> <debug_req_reply_dir>     (saves request and reply messages in files)
>>>
>>>> See
>>>> http://boinc.berkeley.edu/trac/wiki/ProjectOptions#Schedulerdebugging
>>>
>>>> -- David
>>>
>>> _______________________________________________
>>> 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