Chris Goffinet removed another magic number by making the TCP listen backlog runtime configurable here:
http://github.com/lenn0x/memcached/commit/7d010a855ecccf4960fce4ffa4bd79d686130ca6 Before this change, memcached had a hard-coded backlog of 1024. It's possible to exceed 1024 with a connection burst from a large number of clients at which point some clients will see connection refused. Some such large installations have pushed this up to 4096 or so, while others might appreciate saving some kernel memory by keeping it low and having fewer bursty connection bringups. Does anyone have an opinion on this? My only question is whether it should be a compile-time option or a runtime option.
