Rick Jones wrote:
It appears that I can (almost?) completedly deadlock a TCP connection
when
using TSO on 2.6.13.2. My test involves trying to send 200Mbps
between two
interfaces. With TSO enabled (ie, the default for e1000 NICs), in
less than
1 minute, there are no more packets transmitted, though it does run
nice and
fast untill it completely breaks :P
Only 200 Megabits per second?
I didn't ask it to run faster..and it was full duplex to the same machine,
so about 800Mbps total across the bus...
(Yes, this kernel has my send-to-self patch and other of my patches applied.)
This is a single TCP connection?
Yes.
I disabled TSO with ethtool, and now it seems to be working reliably.
Does a new TCP connection "work" or does communication remain hung?
If I stop and start it will start running again..untill it hangs again.
It is likely I am dropping a few packets at this speed, as I did see
a few OOM errors when trying to allocate skbs (this is on a machine with
2GB of RAM, but that is another issue).
It appears easily reproducible..or at least it was yesterday :)
Ben
--
Ben Greear <[EMAIL PROTECTED]>
Candela Technologies Inc http://www.candelatech.com
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
