Repost of a previous message. I'd like to rephrase: Does anybody know if there is some technical necessity (other than memory pressure on the kernel's tcp subsystems) to limit squid to using only 64KB TCP windows? On the client side - LAN with relatively low RTTs - this seems sensible. On outgoing connections - the internet, possibily high or very high RTTs - this seems suboptimal to achieve high thruputs, esp. with larger files. > why is configure limiting the tcp receive buffer size per default to only > 64K on Linux (I know of the setting "tcp_recv_bufsize")? > > Wouldn't it be wise to let the kernel choose the appropriate sizes (because > it only knows about memory pressure or not etc), also with respect to recent > TCP implementation enhancements? Cheers, Ronny
