The google/t-mobilve video is a good perspective on the problem:
The transparent proxies where deployed for a reason. Why for example
could the same throughput goals not easier be achieved with explicit
proxies, eg: http proxies ? Some work from ART that could fix this ?

Given how we know that transparency is a problem, why not try to solve that
by offering options for explicitly visible proxies at transport layer,
eg: with PLUS. Could easily (IMHO) include diagnostics, ot-in or opt-out
options. Use case for example: do not use the proxy when the app thinks
it can do it better itself.

Btw: Diagnostics isn't only bad with most current transparent middleboxes
It eqully worse because of the end-to-end principle. Whats my best option
to localize throughput problems on my oh so wonderfully transparent internet
path - so i can complain to the right ISP ? Pathchar ??? 

Justs to keep things in perspective..

On Fri, Aug 05, 2016 at 01:49:26PM -0700, Yuchung Cheng wrote:
> FWIW Tmobile and Google talked about an experiment to bypass Tmo proxies on
> YouTube traffic in the Velocity conference. It was recorded at
> https://www.youtube.com/watch?v=G_SMCYu7qSw
> (For the impatient, you can skip the first 11 mintues)

On Fri, Aug 05, 2016 at 01:55:58PM -0700, Joe Touch wrote:
> On 8/5/2016 1:21 PM, Fred Baker (fred) wrote:
> >
> >> On Jul 27, 2016, at 11:51 AM, Joe Touch <[email protected]
> >> <mailto:[email protected]>> wrote:
> >>
> >> Sure - my point is that the term "transparent proxy" is common, and
> >> ALL such animals break TCP semantics *by design*.
> >
> > From my perspective (which is always right), the right way to fix
> > these middle boxes is to let users have trouble with them and complain.
> I agree, but users don't always know where to direct their ire.
> 
> > Someone that installs idiot-ware deserves the cost of fixing it.
> 
> They deserve many costs - the cost of lost revenue while they use it,
> the cost of fixing it, etc.
> 
> The trick (as always) is to put the pain where the power lies to solve
> the problem.
> 
> Joe

Reply via email to