PRUS_MORETOCOME is indeed too complex to solve the microcosmic problem of writes between 100 and 208 bytes; however, it solves the more general problem of the Nagle/MTU interaction even when the MTU is larger than a cluster (e.g. loopback, ATM, FDDI, etc). Try the atomic patch (and remove PRUS_MORETOCOME) with writes of 2049-2256 bytes on the loopback interface. Same with LOCAL-domain sockets (with the uipc_usrreq.c patch I sent you).
Bill To Unsubscribe: send mail to majord...@freebsd.org with "unsubscribe freebsd-hackers" in the body of the message
