Date: Sun, 29 Sep 1996 12:31:24 -0400 (EDT) From: Howard Goldstein <hgoldste@news1.mpcs.com> To: FreeBSD-gnats-submit@freebsd.org Subject: kern/1689: TCP extensions throttles throughput Message-ID: <199609291631.MAA06428@news1.mpcs.com> Resent-Message-ID: <199609291640.JAA02387@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
>Number: 1689 >Category: kern >Synopsis: TCP extensions throttles distant connections >Confidential: no >Severity: serious >Priority: high >Responsible: freebsd-bugs >State: open >Class: sw-bug >Submitter-Id: current-users >Arrival-Date: Sun Sep 29 09:40:02 PDT 1996 >Last-Modified: >Originator: Howard Goldstein >Organization: MPC Systems >Release: FreeBSD 2.1.5-RELEASE >Environment: net.inet.tcp.rfc1323=1 >Description: TCP connections between distant (155-210ms ping times) 2.1.0R, 2.1.5R or both systems, as for example two FreeBSD systems over a dialup CSLIP line, with net.inet.tcp.rfc1323=1 (default) results in 3 to 4 sec delays between transmission of subsequent TCP packets even though the dialup line is solid and tcpdump reveals correct window ACKs making a timely return. Topographically distant remote non-dialin http users also reported poor performance. The severity and priority of this problem would have been ranked non-critical/low had the shipped default for TCP extensions been set to off. >How-To-Repeat: Ensuring that both systems have rfc1323=1, telnet to the echo socket of distant system. Distant does not mean ethernet RTT times, the RTT must be in the range given above. Observe the first echoed packet comes back quickly but all subsequent packets suffer 3-5 second delay. Other TCP sockets suffer from same delay but it is most easily observed and tcpdumped with an echo socket. Anyone who needs to reproduce this but can't find a system topologically distant enough should email me or talk hg@cally.south.mpcs.com and I'll turn on a telnettable system with the deadly extension enabled. >Fix: Work around is inclusion of net.inet.tcp.rfc1323=0 in the rc for all FreeBSD systems. (NB., there continue to be RTT computation time problems, problems discussed in a series of postings by Matthew Dillon around 8 Mar 96 on the hackers list but not GNATed, that have not been corrected) >Audit-Trail: >Unformatted:
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199609291631.MAA06428>