Comment 42 for bug 213081

Brian Burch wrote:
> Yes, I understand what is required. Leave it with me for now and I'll do
> some traces on intrepid over the weekend. I will let you know what I've
> found asap. Once I have a good grasp of the protocol issues, then it
> will be time to start looking at the code. (After 15 years with C and
> C++, I jumped ship to java as soon as it was released and never looked
> back... occasionally that experience comes in useful).

I ran a trace of a 220K print file using the Intrepid alpha 4 kernel,
2.6.26-5.17-generic. It took a long time - cups said it was 220K, but
actually transmitted twice that much. It also transmitted datagrams with
only 512 bytes, but I feel these are cups issues, not kernel tcp/ip.

The tcp session appears to be satisfactory. There were some timeouts,
retransmissions, duplicate acks, window closures, etc. All these
"normally abnormal events" were handled in a reasonable manner.

I conclude that the intrepid kernel with Ilpo's fix is correct:- when
sending significant data to remote devices that have primitive tcp/ip
stacks.

I will research the fixes in intrepid to see which ones we should
back-port, but this might take me some time...

Regards,

Brian