[concurrency-interest] Thread subclass check locking is a bottleneck
gregg at cytetech.com
Mon Feb 16 17:59:58 EST 2009
David Holmes wrote:
> There is no need to lock the class at all. The check is a reflective check
> for the existence of an overridden set/getContextClassLoader method - there
> are no concurrency issues there (the answer can never change for a given
> class). The synchronization is needed to update the table, nothing more.
Okay, that's good, so it really should just be a concurrent map with the data
race on update, no problem! This seems to be a big barrier in my application
right now. Was this a recent change? I haven't been seeing this kind of
contention in my thread pool which has been active in this application for about
10 years now.
More information about the Concurrency-interest