Parallel Chunk Loops


I played around with the Parallel Chunk Nodes and have some problems understanding what they are actually doing.

What I’m trying to achieve is getting a better CPU utilization when calculating some Image Segment Features. So all I have to do is putting the Parallel Chunk Start/End Nodes before/after the Image Segment Features Node right ??

At least that’s what I’ve done and I encountered a better CPU utilization running the Image Segment Features Node but after that, the Parallel Chunk End node stays a LONG time at about 90% which I’m not really understanding because the heavy calculations are done in the Image Segment Features Node ?!


Kind regards,


​Hi Flo,

​If you would remove the BitMasks using a ColumnFilter the Parallel Chunk Loop End should be faster. Reason: The Parallel Chunk Loop End Node copies all the data coming from the individual branches which were created to parallelize the process and reassembels them into one table.

Copying all the individual BitMasks currently is somehow slow, so if you don't need them, you can ​simply remove this column and the Loop End should be faster. There exists one option in the File -> Properties -> KNIME -> Image Processing which is the maximum file size. Default it is zero, which means each object (Bitmask) goes into its individual file. ​If you increase the value to lets say 2000mb, then the copying will also be faster, but it has some ugly side-effects, i.e. you can't sort tables in the table view if the table contains images. This is due to a bug in KNIME core which hopefully will be resolved with KNIME 2.11.

​I hope this helps,


That helped a lot thank you :)