Task #9329 (closed)
Opened 11 years ago
Closed 11 years ago
Profile the OME-TIFF writer
| Reported by: | mlinkert | Owned by: | mlinkert |
|---|---|---|---|
| Priority: | minor | Milestone: | OMERO-4.4.4 |
| Component: | Bio-Formats | Version: | n.a. |
| Keywords: | n.a. | Cc: | sbesson |
| Resources: | n.a. | Referenced By: | n.a. |
| References: | n.a. | Remaining Time: | n.a. |
| Sprint: | n.a. |
Description
We've had reports that writing OME-TIFF files (especially through MATLAB) is slow; Sebastien say that it takes ~45seconds to write 1000 256x256 planes using bfsave.m.
While this isn't terrible, there is probably room for improvement, so it would be nice to run the OME-TIFF writer through a profiler (I've been using JRat) and see what the bottlenecks are.
Change History (5)
comment:1 Changed 11 years ago by sbesson
comment:2 Changed 11 years ago by mlinkert
- Milestone changed from OMERO-4.4.x to OMERO-4.4.1
comment:3 Changed 11 years ago by mlinkert
As far as I can see, the performance issues can be reasonably addressed with this commit:
https://github.com/melissalinkert/bioformats/commit/325f6bd93381c6774985dbcc8d8191286da2caf9
...which is a change to bfsave.m, and not a change to the underlying TIFF writing logic. Sebastien, can you please confirm that that commit works and is dramatically more performant?
comment:4 Changed 11 years ago by sbesson
Tested with 1000 256x256 images: bfsave function executing ~3X faster than previously.
Also pushed a commit implementing a test case for this ticket:
https://github.com/sbesson/bioformats/commit/23497e6317f09049e833c546c67f7b567e892db2
comment:5 Changed 11 years ago by mlinkert
- Resolution set to fixed
- Status changed from new to closed
PR from Sebastien merged into my sprint2-bug-fixes and BioFormats-Matlab updated to reflect the fact that anyone writing planes sequentially should really use the setWriteSequentially method in IFormatWriter.
As the first step of a performance test, wrote a script comparing the execution time of bfsave/Matlab imwrite (https://gist.github.com/3088736)
Ratio of execution times varies between 2-10 depending on image size. This can certainly be improved in the Java code as I don't expect the overhead from copying arrays to account for such a large ratio (cf estimated overhead time ratio ~1.5-2.5 from bfopen.m).