Date: Mon, 14 Aug 2000 14:09:01 +1000 From: Peter Jeremy <peter.jeremy@alcatel.com.au> To: current@FreeBSD.ORG Subject: Re: Slow ssh throughput with -current Message-ID: <00Aug14.140905est.115211@border.alcanet.com.au> In-Reply-To: <00Aug10.153604est.115242@border.alcanet.com.au>; from peter.jeremy@alcatel.com.au on Thu, Aug 10, 2000 at 03:36:01PM %2B1000 References: <00Aug10.153604est.115242@border.alcanet.com.au>
next in thread | previous in thread | raw e-mail | index | archive | help
On 2000-Aug-10 15:36:01 +1000, Peter Jeremy <peter.jeremy@alcatel.com.au> wrote: >I just noticed a really peculiar network slowdown with -current from >about a week ago. The machine has two NICs - an old 16-bit SMC card >(ed0) running 10baseT and a PCI SMC card (tx0) running 100baseTX full >duplex. > >When I scp (either ssh-1.2.27 or OpenSSH) to a remote machine via ed0, >I get a throughput of ~8KB/sec. When I go via tx0, I get ~3MB/sec >(CPU limited). If I use ftp via ed0, I get wire speed (just over >1MB/sec). Having done some more tests, and a bit more digging, I've found the culprit is net.inet.tcp.newreno. When set to 1 (the default), ssh (but not FTP) connections to hosts not on the same subnet (as determined by in_localaddr()) have an effective window size of 1 MSS. Unfortunately, I can't see any obvious reason why initialising maxburst to 4 results in only a single packet on each call to tcp_output() for ssh connections. Peter To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?00Aug14.140905est.115211>