TCP Retransmission & lost segments problem under Linux but not under XP

nx5000 · 12-16-2007, 05:38 AM

Go on with what unSpawn said.

But meanwhile, I have one very quick question:

Did you try the mtu stuff? It takes about 20seconds to test!
Check it's correctly applied by typing ifconfig and verifying the mtu has changed.

There has been bad interactions between recent Linux kernels and not standard webservers.

Kernel bug maybe.
Router bug ?
Webserver bug?
Chair-keyboard Interface bug? ;p

Seriously, at this step, you have no information at all to make any conclusion.

unSpawn · 12-16-2007, 06:11 AM

Quote:

Originally Posted by nx5000

There has been bad interactions between recent Linux kernels and not standard webservers.

Any specific things you got pointers to?

nx5000 · 12-16-2007, 06:36 AM

Oh my, this will give me work for the whole day grepping changelogs and mailinglists.
I'll edit the post when I find it. I think it was around 2.6.20-2.6.22

Here's an example of broken routers (but it's not the one I was looking for)
http://kerneltrap.org/node/6723

unSpawn · 12-16-2007, 11:24 AM

Quote:

Originally Posted by nx5000

this will give me work for the whole day grepping changelogs and mailinglists

Thanks for the URI. Not really necessary to grep all sorts of stuff though, I thought maybe you had some stuff jotted down somewhere.

debuser123 · 12-16-2007, 02:49 PM

Quote:

Originally Posted by nx5000

Here's an example of broken routers (but it's not the one I was looking for)
http://kerneltrap.org/node/6723

But it was EXACTLY what I was looking for! TCP window scaling was the problem! Clicking through links from the page you directed I came across a pretty concise description of the problem (TCP window scaling and broken routers[corbet]). As soon as I disabled window scaling (echo 0 > /proc/sys/net/ipv4/tcp_window_scaling, I was instantly able to view the web page).

This is what is happening when window scaling is enabled:
me: (syn) window=5840, scale: 5
them: (syn-ack) window=16384, scale: 0
me: (ack) window=5856 (scaled)
me: (psh-ack) -http_get- window=5856 (scaled)
me(retry): (psh-ack) -http_get- window=5856 (scaled)
them: (ack) window=65084 [previous seg lost]
15 seconds...
them: (rst)

and from that article:

Quote:

[...] some routers on the net are rewriting the window scale TCP option on SYN packets as they pass through. In particular, they seem to be setting the scale factor to zero, but leaving the option in place. The receiving side sees the option, and responds with a window scale factor of its own. At this point, the initiating system believes that its scale factor has been accepted, and scales its windows accordingly. The other end, however, believes that the scale factor is zero. [...] In many cases, the small window can cause no packets to be transmitted at all, breaking TCP between the two affected systems entirely.

So the consensus is...there's a bad router.....somewhere. Just having to remember to disable window scaling for this site (or any sites where I can connect and the browser's status is stuck at "waiting for server.com..." until it the connection gets ultimately reset) is something I can live with.

(Rebooting, wanna see what a pcap looks like under Windows)
Edit: Windows does not use scaling. Packets I initiated had a window of 64K and packets received, a window of 16K. When I disabled scaling under 2.6, all my packets had a window of 5840 while the remote host, 16K).

Thanks you guys, really appreciate it.

and PS: I did try that MTU fix but that problem still arose.

unSpawn · 12-16-2007, 02:55 PM

Well done nx5000 :-]

debuser123 · 12-16-2007, 03:28 PM

See nx5000, your advice wasn't crap! I can now view this porn site thanks to you & unSpawn!JUST KIDDING!

nx5000 · 12-16-2007, 04:34 PM

Quote:

Originally Posted by debuser123

See nx5000, your advice wasn't crap! I can now view this porn site thanks to you & unSpawn!JUST KIDDING!

Great for you :]

I'll play loto tomorrow.