You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
After fixing the expression matching performance improvement from #206, CPU utilization during pass 2 node handling dropped, but node processing time remained about the same (4 minutes on a 64 core machine). Digging into this it turns out to be that ICU4J transliteration is synchronized across threads. Setting --transliterate=false removes the thread contention and reduces overall processing time by 2-3 minutes.
See if there's a way to do ICU4J name transliteration without requiring global synchronization.
The text was updated successfully, but these errors were encountered:
After fixing the expression matching performance improvement from #206, CPU utilization during pass 2 node handling dropped, but node processing time remained about the same (4 minutes on a 64 core machine). Digging into this it turns out to be that ICU4J transliteration is synchronized across threads. Setting
--transliterate=false
removes the thread contention and reduces overall processing time by 2-3 minutes.See if there's a way to do ICU4J name transliteration without requiring global synchronization.
The text was updated successfully, but these errors were encountered: