Re: The file that kills linux tcp

Robin Cutshaw (robin@intercore.com)
Wed, 5 Feb 1997 09:27:29 -0500

Alan Cox writes:
> > In that case, it would depend on the IP address of both the sending
> > and the receiving ends... which is just tricky enough that it could be
> > true.
>
> No you only need about 200 bytes of the right pattern to derail faulty
> 100baseT.
>

I've got a similar problem. My lab net has several hosts (linux-ix86,
linux-axp, freebsd-216, freebsd-current, freebsd-22beta, unixware,
solaris-x86, solaris-sparc, sunos-sparc, netbsd-ix86). The two linux
boxes (ix86 and axp) running 2.0.28 both will have tcp freezes (a rcp
stalls in the middle of the transfer) when copying large distributions
(100MB+) to a server. These hosts are on a 100baseT hub. Only the
linux boxes show this problem. Wouldn't this point to a linux
(maybe de driver) problem. All of the ix86/axp boxes use the DE500
(21140) cards.

robin

-- 
----
Robin Cutshaw         internet: robin@interlabs.com robin@intercore.com
Internet Labs, Inc.   BellNet:  404-817-9787


    "Time is just one damn thing after another" -- PBS/Nova
----

--