http://bugs.winehq.org/show_bug.cgi?id=59765 --- Comment #7 from mcsordas <csoresz2@gmail.com> --- Hello, We continued testing on a live RTSP-over-TCP workload with 43 concurrent streams. Without any application-side optimizations, wineserver CPU was around 95%. We then tried a few application-side workarounds to reduce the number of real Winsock calls reaching Wine. These are not fully clean or generally reliable solutions, but the results are interesting: Reducing real recv() calls with application-side buffering lowered wineserver CPU from ~95% to ~75%. About 80% of recv() calls were served from the application buffer. Then reduced real select() calls by using application-side readiness emulation when buffered data was already available. About 60% of select() calls were avoided/emulated. This lowered wineserver CPU further to ~63%. Keeping those changes and adding a small 1 ms delay before the remaining real select() calls lowered wineserver CPU to ~50%. This seems to support the idea that the bottleneck is strongly tied to the number of socket I/O / readiness operations that reach wineserver. However, these are application-side hacks involving buffering, readiness emulation, and throttling. They work in our tests, but they are not a clean or generally safe solution. -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.