I'm developing a BOINC project to calculate Spectral Energy Distributions for Astronomy images, but need some advice on tracking down an annoying little bug. We are currently testing the application out using AWS to host it. The project uses the BOINC wrapper to wrap some Fortran code. Essentially it processes a number of pixels one at a time. Credit is assigned on the number of pixels process as it uses a simple "brute force" model fitting approach.

The project is up and running on 32/64 bit Windows, OS X, and 32bit/64bit Linux. Occasionally I am getting absolutely stupid cpu_time values as shown below:

mysql> select userid, name, cpu_time,claimed_credit, granted_credit from result;
+--------+----------------------------------+--------------+----------------+----------------+
| userid | name                             | cpu_time     | claimed_credit | granted_credit |
+--------+----------------------------------+--------------+----------------+----------------+
|      1 | GI3_050001_NGC628_0001__wu31_0   |     677466.6 |              0 |             27 |
|      1 | GI3_050001_NGC628_0001__wu31_1   |       681845 |              0 |             27 |
|      1 | GI3_050001_NGC628_0001__wu31_2   |     680222.4 |              0 |             27 |
|      1 | GI3_050001_NGC628_0001__wu9_0    |       698966 |              0 |              0 |
|      3 | GI3_050001_NGC628_0001__wu12_2   | 1.218908e254 |              0 |              1 |
|      1 | GI3_050001_NGC628_0001__wu377_0  | 1.553102e254 |              0 |             39 |
|      1 | GI3_050001_NGC628_0001__wu108_0  | 1.553102e254 |              0 |             49 |
|      4 | GI3_050001_NGC628_0001__wu13_1   |  2.71667e254 |              0 |             34 |
|      4 | GI3_050001_NGC628_0001__wu707_1  |  9.07193e216 |              0 |              2 |
|      4 | GI3_050001_NGC628_0001__wu96_0   |     337490.8 |              0 |             10 |
|      1 | GI3_050001_NGC628_0001__wu96_1   |     322856.6 |              0 |             10 |
|      1 | GI3_050001_NGC628_0001__wu96_2   |     326189.8 |              0 |             10 |
|      1 | GI3_050001_NGC628_0001__wu74_0   |     305999.3 |              0 |             10 |
|      1 | GI3_050001_NGC628_0001__wu74_1   |     306578.1 |              0 |             10 |
|      4 | GI3_050001_NGC628_0001__wu588_0  |     934082.2 |              0 |             49 |
|      4 | GI3_050001_NGC628_0001__wu588_1  |       929387 |              0 |             49 |
|      4 | GI3_050001_NGC628_0001__wu588_2  |       918652 |              0 |             49 |

Some of the cpu numbers are greater than 1e216 - which is a silly number. Could someone point me in the right direction to debug this?

The odd thing is the same client can run the job twice and one run is fine, whilst the next gives the stupid values. I've seen it happen on OS X, 32 & 64 bit windows. Not linux yet, but I only have one test client at the moment there

Thanks in advance
Kevin

This is solved in the latest wrapper.
yoyo

---

Germany biggest distributed computing community Rechenkraft.net