mirror/OCRmyPDF
1
0
Fork 0
mirror of https://github.com/ocrmypdf/OCRmyPDF.git synced 2026-05-05 05:05:44 -04:00
Code Issues Packages Projects Releases Wiki Activity

Use at most 3 Tesseract threads

Browse Source 
Based on a user suggestion and
tesseract-ocr/tesseract#2611, I reviewed thread limits and found that
thread limit of 3 is still beneficial, but not 4.

> time env OMP_THREAD_LIMIT=2 tesseract omp4.png stdout >/dev/null
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 143
116.67user 1.67system 1:26.26elapsed 137%CPU (0avgtext+0avgdata 356752maxresident)k
2213inputs+0outputs (18major+131059minor)pagefaults 0swaps
> time env OMP_THREAD_LIMIT=3 tesseract omp4.png stdout >/dev/null
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 143
136.89user 1.63system 1:19.56elapsed 174%CPU (0avgtext+0avgdata 356784maxresident)k
821inputs+0outputs (0major+131080minor)pagefaults 0swaps
> time env OMP_THREAD_LIMIT=4 tesseract omp4.png stdout >/dev/null
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 143
161.31user 1.51system 1:18.80elapsed 206%CPU (0avgtext+0avgdata 356632maxresident)k
8477inputs+0outputs (12major+131074minor)pagefaults 0swaps
> time env OMP_THREAD_LIMIT=8 tesseract omp4.png stdout >/dev/null
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 143
160.30user 1.62system 1:18.01elapsed 207%CPU (0avgtext+0avgdata 356640maxresident)k
821inputs+0outputs (0major+131078minor)pagefaults 0swaps
This commit is contained in:
James R. Barlow
2019-09-20 17:11:29 -07:00
parent de61530d4d
commit 78e8bf9cbf
1 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
src/ocrmypdf/_sync.py
View File

@@ -225,15 +225,15 @@ def exec_concurrent(context):
if max_workers > 1:
context.log.info("Start processing %d pages concurrent", max_workers)
# Tesseract 4.0 is multithreaded, and we also run multiple workers. We want to
# avoid the situation where we end up trying to run NxN jobs on N CPU cores,
# as that gives poor performance. Performance testing shows we're better off
# Tesseract 4.x can be multithreaded, and we also run multiple workers. We want
# to manage how many threads it uses to avoid creating total threads than cores.
# Performance testing shows we're better off
# parallelizing ocrmypdf and forcing Tesseract to be single threaded, which we
# get by setting the envvar OMP_THREAD_LIMIT to 1. But if the page count of the
# input file is small, then we allow Tesseract to use threads, subject to the
# constraint: (ocrmypdf workers) * (tesseract threads) <= max_workers and limiting
# Tesseract to 4 threads.
tess_threads = min(4, context.options.jobs // max_workers)
# constraint: (ocrmypdf workers) * (tesseract threads) <= max_workers.
# As of Tesseract 4.1, 3 threads is the most effective on a 4 core/8 thread system.
tess_threads = min(3, context.options.jobs // max_workers)
if context.options.tesseract_env is None:
context.options.tesseract_env = os.environ.copy()
context.options.tesseract_env.setdefault('OMP_THREAD_LIMIT', str(tess_threads))