Hello! > buffer memory allocation, counted in byte. Right? Almost. These numbers are counted in pages. > microseconds. Now the write takes 15ms (for the 14kB requests), and 50ms > (for the 48kB requests) - all the while blocking the operation of the > nice big poll loop server. No, the reason is not here. The reason is that _default_ value for TCP sndbuf (tcp_wmem, middle value) is strongly decreased. It is 16K now, so that the first write() of > 16K (really less due to overhead) blocks. I do _NOT_ advise to increase it. If you build real internet web server, you probably even will have to decrease it to 4K or so. If you want to restore old behaviour (which resulted in memory collapses, unless web server was scoped to LAN) you may increase it to old value: 65535. Alexey - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org
