Much longer running tasks
log in

Advanced search

Message boards : Windows : Much longer running tasks

Author Message
Supracore
Send message
Joined: 23 Jun 17
Posts: 2
Credit: 9,749,508
RAC: 0
Message 467 - Posted: 10 Sep 2017, 11:50:56 UTC

I've seen that some tasks run much longer than the other.
Like this one:
Stderr Ausgabe

<core_client_version>7.8.2</core_client_version>
<![CDATA[
<stderr_txt>
Command line arguments recieved:

Atomic ensemble calculation time: 22.6824 s

Selected OpenCL device:
GPU: Ellesmere
Number of compute units: 36
GPU clock rate: 1303 MHz
Theoretical peak performance: 6004 GFLOPs

Histogram calculation time: 2071.03 s
1D pattern calculation time: 0.2028 s
Total calculation time: 2096 s
11:00:37 (3248): called boinc_finish(0)

</stderr_txt>
]]>
____________

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 321
Credit: 103,382
RAC: 0
Message 468 - Posted: 10 Sep 2017, 17:51:07 UTC - in response to Message 467.

I've seen that some tasks run much longer than the other.

Thank you for reporting this!
For a fixed crystallite size, the execution time of the WU is proportional to the square of the atomic density of the structure. The distribution of the processed COD structures on their atomic density look like this. The average atomic density is 0.091 A^(-3). Based on the stdout file, the atomic density of the COD structure from your example is 0.53 A^(-3), which is 5.8 times greater than the average. This is an abnormal atomic density. Probably, the original CIF file contain an error.

Profile PDW
Send message
Joined: 17 Nov 16
Posts: 20
Credit: 360,217,503
RAC: 7,327
Message 469 - Posted: 10 Sep 2017, 18:14:47 UTC - in response to Message 468.

Hi,

I've just found one that has been running for 1 hour and 20 minutes so far on a GPU. It's coming up to 88% complete, others who have run it got similar run times.

This is the task: http://xansons4cod.com/xansons4cod/result.php?resultid=19139612
I expect it to finish in the next 15 minutes.

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 321
Credit: 103,382
RAC: 0
Message 470 - Posted: 10 Sep 2017, 20:26:41 UTC - in response to Message 469.

Hi,

I've just found one that has been running for 1 hour and 20 minutes so far on a GPU. It's coming up to 88% complete, others who have run it got similar run times.

This is the task: http://xansons4cod.com/xansons4cod/result.php?resultid=19139612
I expect it to finish in the next 15 minutes.

Hi PDW,

Thanks! It seems, all these structures are from the same book published in 1951, http://www.crystallography.net/cod/result.php?CODSESSION=dg7luthf8fh15s4n0qog0ecd6pq4evj0&page=0&count=200&order_by=file&order=asc. Most of them are OK, but some have problems with symmetry operations. Namely, the symmetry equivalent positions that must be degenerate are not, and the app believes that the structure contains more atoms than it actually contains.

Profile PDW
Send message
Joined: 17 Nov 16
Posts: 20
Credit: 360,217,503
RAC: 7,327
Message 471 - Posted: 10 Sep 2017, 20:31:41 UTC - in response to Message 470.

Is this something that needs fixing and can be fixed ?

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 321
Credit: 103,382
RAC: 0
Message 472 - Posted: 10 Sep 2017, 21:09:32 UTC - in response to Message 471.

Is this something that needs fixing and can be fixed ?

Technically, it can be fixed without modifying the input data. Now, the two atomic positions are considered degenerate if the following condition is met:
((x0 - x1) < 0.001) && ((y0 - y1) < 0.001) && ((z0 - z1) < 0.001),
where (x0,y0,z0) and (x1,y1,z1) are their fractional coordinates.
This is a very strict condition, but it works fine for 99.9% of COD entries. I can relax this condition but I do not know the proper value for the tolerance yet.

Profile PDW
Send message
Joined: 17 Nov 16
Posts: 20
Credit: 360,217,503
RAC: 7,327
Message 473 - Posted: 10 Sep 2017, 21:17:17 UTC - in response to Message 472.

Okay, you can make the holes in the net bigger but not sure how much bigger to make them in case real fish get through ?

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 321
Credit: 103,382
RAC: 0
Message 474 - Posted: 11 Sep 2017, 11:33:45 UTC - in response to Message 473.

Okay, you can make the holes in the net bigger but not sure how much bigger to make them in case real fish get through ?

Yes, this metaphor is the right one). However, when I visualized the structure in Mercury, I saw that the problem can’t be solved that simple, because the atoms that "pretend" to share the same atomic position are the atoms of different chemical elements. So, the app still has to decide (somehow) which chemical element should occupy this particular position, and this is much more complicated problem. For now, we have about 100 structures (out of ~ 250 000 processed) with atomic density higher than that of a diamond. It’s possible to check them manually and report the possible errors to the COD maintainers. I’ll try to do that when the calculations will be completed.

Profile PDW
Send message
Joined: 17 Nov 16
Posts: 20
Credit: 360,217,503
RAC: 7,327
Message 477 - Posted: 12 Sep 2017, 8:12:24 UTC - in response to Message 474.

Okay, thanks for the update.
That Mercury tool looks pretty nifty !

Win10
Send message
Joined: 10 Jul 17
Posts: 4
Credit: 47,870,721
RAC: 5
Message 480 - Posted: 12 Sep 2017, 23:38:23 UTC
Last modified: 12 Sep 2017, 23:38:50 UTC

here is another very long WU:
http://xansons4cod.com/xansons4cod/workunit.php?wuid=9311925

Vlad
Project administrator
Project developer
Project tester
Project scientist
Help desk expert
Send message
Joined: 26 Oct 16
Posts: 321
Credit: 103,382
RAC: 0
Message 481 - Posted: 13 Sep 2017, 11:49:26 UTC - in response to Message 480.

here is another very long WU:
http://xansons4cod.com/xansons4cod/workunit.php?wuid=9311925

Thanks! This one was completed but I found another one still running, so I killed it. It appears that currently no very long WUs left running (at least for the structures from this list).

Message boards : Windows : Much longer running tasks


Main page · Your account · Message boards


© 2020 Vladislav Neverov (NRC 'Kurchatov institute'), Nikolay Khrapov (Institute for Information Transmission Problems of RAS)