[Developers] Parallelization of computations

[Kristofer Tingdahl]

Colin Hawke wrote:
> Hi Kristofer,
>
> I have a quick query to this. I have a 5 node, 40-core cluster (each node has two quad-core cpu's). In OD some processing steps such as generating/filtering steering cubes seem to utilise the cores very nicely. However other jobs such as making a structurally orientated filtered volume (using Volume Statistics with Full Steering) only one core (out of 8 !) seems to get utilsed in each node - any idea why this is so ?
>
> Thanks
>
> Colin
>   
Dear Colin,

There can be a number of reasons for this: First of all, the full 
steering is implemented in a way that does not use parallel 
computations. This is something we may want to look into. The time spend 
in full steering is however relatively minor, so the impact should be 
limited. Another reason may be data i/o, i.e. how fast the nodes may get 
the data. The steering cube generation is very good in parallel 
processing, since the i/o is limited and it's a lot of calculations. 
Filtering operations are often cheap in terms of processing, but the i/o 
is often more extensive.

That is at least our experience when processing attributes.

Best Regards,

Kristofer

-- 
Dr. Kristofer Tingdahl
dGB Earth Sciences
1 Sugar Creek Center BLVD #935; Sugar Land, TX 77478; USA
+1 281 652 5318