On Mon, 26 Nov 2001, Manuel Bouyer wrote:

=> > Also, wouldn't a wrong duplex setting affect speeds in both directions,
=> > and to the same subnet, too?  The problem I am experiencing is only when
=>
=> No, as the link behavior isn't symetric in this case.
=> things will be worse on the local subnet if the remote machine is
=> full-duplex. However if it's half-duplex you'll see almost no
=> collisions on your link.
=>
=> What driver do you use ?

The lance ethernet driver:

NetBSD 1.5Y (CHUMBY) #2: Wed Nov 21 11:44:53 EST 2001
    paul@chumby:/usr/src/sys/arch/alpha/compile/CHUMBY
DEC 3000 - M300LX, 125MHz, s/n
8192 byte page size, 1 processor.
total memory = 112 MB
[...]
ioasic0 at tc0 slot 5 offset 0x0: slow mode
le0 at ioasic0 offset 0xc0000: address 08:00:2b:3d:d7:b8
le0: 32 receive buffers, 8 transmit buffers

=> Does netstat -i show errors on the interface ?

No errors.

Name  Mtu   Network       Address              Ipkts Ierrs    Opkts
Oerrs Colls
le0   1500  <Link>        08:00:2b:3d:d7:b8  6075433     0  8829331
0  4521



Actually, I think I just solved the problem.  I noticed that one of the
differences between my 1.5 DEC3000/300 and 1.4 DECstation 5000/240 is
that net.inet.tcp.cwm is set to 1 on the former and 0 on the latter.
When I set net.inet.tcp.cwm to 0 on the NetBSD 1.5 box, the problem went
away.  When I set it to 1, FTP speeds went straight down to ~7 KB/sec
when sending from the 1.5 machine to another subnet.

I guess net.inet.tcp.cwm defaults to 1 nowadays.

Cheers,

Paul.

e-mail: paul@gromit.dlib.vt.edu

"Without music to decorate it, time is just a bunch of boring production
 deadlines or dates by which bills must be paid."
        --- Frank Vincent Zappa