Dreadful xfer rates and TCPv4 bad checksum in 2.1.1xx with PPP

David G Hamblen (dave@afr-olt.com)
Thu, 5 Nov 1998 07:26:49 -0500 (EST)

I've been following the kernel digest, and searching for solutions to this
problem, and several recent posts are suggesting that the NIC driver is at
fault, or that its just a debugging message which will go away in 2.2. I
get 1/3 the ftp transfer rate with 2.1.125 and 2.1.127-pre7 compared to
2.0.33. My problem occurs using ppp and serial drivers; so I would think
the only common thing between my problem and the others is the tcp_ipv4
code. Since I've got an old tired serial interface, my guess is that
interrupts aren't being serviced fast enough?

I'm not sure just when these started being a problem, but with 2.1.12x
kernels, I get lots of these messages and my ftp downloads drop from
1.8kb/s with 2.0.33 to 0.6Kb/s with 2.1.125 to 0.3Kb/s with 2.1.127-pre7.
This is with my 14.4 modems on a 30K file on an unloaded ISP. The 2.0.33
kernel runs consistently at 1.8Kb/s. This makes the 2.1.12x kernels
pretty much unusable for email. The syslogd messages file will get a
dozen or so TCPv4 bad checksum messages, while ifconfig shows maybe 3
errors, and 3 frames. I've tried /dev/ttyS0? at 38400 and 19200 with no
variation in performance. The line is connecting at 14,400 (usually -))

When connected at work to the same server on our LAN via eth0 (a 3COM
pcmcia card) these checksum errors rarely appear, which would indicate
either the serial driver or ppp. I tried no compression switches in my
ppp options to no avail.

System: (admittedly mature)

Intel 486 66 Mhz
16450 UART
Brand X (Rockwell chip) internal 14.4 fax/modem on /dev/ttyS01
Hayes Accura 14.4 external fax/modem on /dev/ttys00
ppp-2.3.5
ifconfig 1.35
nettools 1.46

I've tried compiling the serial driver as builtin and as a module, no
difference.

Dave

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/