[Cerowrt-devel] Better results from CeroWrt 3.10.28-16

[Rich Brown]

Dave,

Thanks for looking at this.

> Wow. Mac TCP is certainly getting stuffed at 768KB up.

This sounds like an OSX problem (which I think we always suspected. *Every* network stack needs to have this worked on.)

Does anyone know how to approach Apple about this? I am a registered Apple developer (me, and my 100,000 closest friends). I don't think I have any connections there that could rise above the noise.

> using a wireshark filter of tcp.stream eq 18
> 
> it looks like you are iw3, with 8 bytes of overhead on the wire, and
> although the MSS gets reduced slightly (not enough!)
> the window never gets below 132480. The throughput graph is reasonable
> regardless.... but I have to figure out a way to "see" if the RTO
> timer is firing or not. Thankfully there's an expert handy this
> afternoon.
> 
> I'll setup a duplicate test here on a different OS and see what I see...

I will see if I can get netperf-wrapper installed on my Win7 laptop. That effort will compete for time with my work week, though.

> supplying the rrul *json.gz files would let me do some totals plots....

https://dl.dropboxusercontent.com/u/21403660/RRUL-json.zip

> 
> On Sun, Mar 2, 2014 at 6:28 PM, Rich Brown <richb.hanover at gmail.com> wrote:
>> Dave,
>> 
>> I got the wireshark info. https://dl.dropboxusercontent.com/u/21403660/RRUL-Wireshark.zip
>> 
>> It was my intent to replicate the RRUL -p all_scaled test with both 'auto' and '30ms' for the egress target. I hope I got it correct - both the settings and the wireshark capture.
>> 
>> Rich
>> 
> 
> 
> 
> -- 
> Dave Täht
> 
> Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html