Now that we do not need the new object while holding the highly trafficed
objhead lock, don't preallocate objects. One unfortunate effect of this preallocation, only recently identified: On systems with high number of worker threads and high hitrates, considerable time could elapse between the preallocation of an object to a worker thread and the use of that object. This could make cache-misses use a much larger working set size than really necessary. I do not have access to any benchmarks that show a credible performance difference, but reports are most welcome. git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@3842 d4fa192b-c00b-0410-8231-f00ffab90ce4
Showing
Please register or sign in to comment