Messages in this thread | | | Date | Thu, 17 Dec 1998 11:27:07 -0500 | From | Raul Miller <> | Subject | Re: tcp lossage in 2.1.131 |
| |
Alan Cox <alan@lxorguk.ukuu.org.uk> wrote: > > A few weeks ago, I reported on some systematic tcp packet corruption (when > > talking to netbsd from a recent linux 2.1.*). I still reliably experience > > this with 2.1.131. I don't experience this from a linux 2.0.* kernel > > on a similar (pentium) platform, nor do I experience this from Solaris. > > How do you know ? What tool are you using to verify the checksums, remembering > tcpdump doesnt. It would also be useful to know via a third machine exactly > which network hop the frames changed. Also the media involved in each hop
Actually, I don't know -- and looking deeper it's not what I originally thought.
What I do know: when the netbsd box talks to my 2.1.131 box, 2.1.131 ignores the first copy of each tcp data from the remote machine. Other machines (2.0.* linux and solaris) don't experience this problem (and I presume that if lots of people saw it that I'd have encountered some documentation on the issue).
The netbsd box's tcpdump reports something different for the timestamp on the afflicted packets than what the linux box reports. Then again, there don't appear to be any timestamps on the packets between the netbsd box and a linux machine which doesn't experience this packet lossage.
Also, I just last night took a look at the packets en-route between the two boxes, and a linux 2.0 box in the middle sees the same timestamps on the packets in flight as 2.1.131.
So, my current hypothesis is that it's the netbsd box that's screwed up, but I'd like to turn off whatever the option is that made timestamps show up for connections to netbsd so I can get a decent connection between these two boxes. [If anyone knows enough about netbsd to tell me how to fix the remote machine, I'll pass the information along, but I at least want to understand this well enough that I don't go saying wild and crazy things anymore about this particular subject.]
> > [Last time I posted tcpdumps from both ends of a tcp connection, showing > > bad timestamps on the corrupted packets. Should I post that kind of > > thing again? Something else?] > > I've still got them, I didnt pay it much attention originally.
Well, I no longer think it's packet corruption, but it is a persistent problem.
Thanks,
-- Raul
- 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/
| |