T1 troubles fixed (was HEADS UP: T1 troubles w/ dragonflybsd.org)

Matthew Dillon dillon at apollo.backplane.com
Mon Jul 5 02:44:00 PDT 2004


    I've solved the T1 issues for now.  It turns out that my Cisco was 
    set to default queueing limits (fair-queue 75) and throwing away output
    packets due to the load.  So many drops were occuring that TCP backoff
    was reducing output throughput to 800Kbits/sec instead of staying pegged
    at 1.5MBits/sec.

    My servers have net.inet.tcp.inflight_enable turned on which means,
    theoretically, that the TCP windows will not get out of control if I
    increase the queue limits, so that is what I did.  Now the output
    throughput is pegged at 1.5MBits/sec and data flow seems to be running
    smoothly, with only occassional bursts of drops.

    leaf.dragonflybsd.org users may notice some lag at times, but it should
    mostly be ok.

					-Matt
					Matthew Dillon 
					<dillon at xxxxxxxxxxxxx>





More information about the Kernel mailing list