Hi, I am experimenting with asio in order to replace our current tcp library and stumbled over something weird this morning: The ansynchronous tcp echo server from the examples is much slower (~factor 3) with compiler optimization turned on! http://www.boost.org/doc/libs/1_36_0/doc/html/boost_asio/example/echo/async_... For my test I modified the blocking single threaded tcp echo client to send 1000 identical messages of about 100kB length to the server and display the echo (see attachment). with -O0: ========= time ./src/TcpClient localhost 3000 | wc 1000 3000 104459000 real 0m6.072s user 0m5.564s sys 0m0.508s with -O3: ========= time ./src/TcpClient localhost 3000 | wc 1000 3000 104459000 real 0m21.287s user 0m5.756s sys 0m0.164s The optimization has very little effect on the client, though. When performing several runs, the time needed for the experiments varies a lot with -O3, ranging from about 13 seconds to up to 40 seconds. With -O0, the values do not change much (6-7 seconds). Other optimization switches (-O1 or -O2) behave like -O3. I have not tried to turn on/off the single optimization flags like -ftree-dominator-opts yet. Some system information (please do not hesitate to ask for more): Ubuntu 8.04, 64bit g++ 4.2.3 With other parts of boost I checked, opimization works as expected, sometimes with outstanding results: My date parser using spirit gained almost a factor of ten in performance. Any ideas what might be happening with the server? Thanks and regards, Roland