[Ns-developers] gsoc wrapup

[Sam Jansen]

Yes, we need to understand this and fix if possible. I've had
validation of the 64-bit effort on my TODO list for a long time but
haven't got around to it.

There isn't anything obviously wrong; 64-bit runs fine and produces
good-looking results, but there is something subtle creating some
differences that we need to understand. I was planning on doing this
in the next 2 weeks at the absolute latest.

2008/8/29 Florian Westphal <fw at strlen.de>:
> Mathieu Lacage <mathieu.lacage at sophia.inria.fr> wrote:
>> > As for the nsc stacks, the main problem is that the Linux stacks
>> > that work on both x86 and x86_64 generate different pcap traces
>> > depending on the host architecture (for instance, it will pick different TCP
>> > source ports on x86 vs. _64).
>>
>> oh ? Would you care to elaborate ?
>
> Sorry, I havn't investigated why its different.
> On 2nd glance, the port numbers are the same on 2.6.26 32 and 64 bit
> kernels, but differ on 2.6.18 32 vs. 64 bit builds.
>
> Traces are different in any case, e.g. with 2.6.26 the first 100 packets
> or so are the same, but then you'll see the push flag set in one of the
> traces. (i ran them through tcpdump -r and diff'd the output).
>
> I've put up the traces i get on my box (x86) and on ns-test (x86-64)
> here:
> http://strlen.de/cradle/trace_32_64/
> (please fetch the .bz2 ones; the uncompressed pcaps are for those
> that want to take a look at the traces but don't have the bunzip2
> unpacker)
>
> I'll see if I can find the cause of this.
>
> Florian
>