[e2e] Are we doing sliding window in the Internet?

[Injong Rhee]

On Jan 5, 2008, at 2:42 PM, Joe Touch wrote:

>
>
> Injong Rhee wrote:
> ...
>> I don't see the
>> penalty that users, ISPs, academics and Mr. Joe next door have paid
>> because of its use. I don't see the Internet is being crashed or
>> crumbling down because of CUBIC.
>
> You don't see it because nobody is measuring it or even looking for  
> it.
> Deploying a protocol and not seeing a problem isn't proof that it's  
> not
> causing harm now, and it's not proof it won't cause harm in certain
> deployment scenarios.
>
> If "works most of the time for most people" were sufficient, we could
> pare the TCP state machine down by 1/3, and cut out all options
> including SACK.
>
> Yes, not causing harm right now is a start, but it's nowhere near  
> the end.
>
> Joe
>


Yes. There are no reported cases of harm in real production networks.  
Just yet.

But if you have seen my work on the evaluation of CUBIC/BIC (http:// 
netsrv.csc.ncsu.edu/twiki/bin/view/Main/BIC#Experimental_Results),  
they are quite extensive.  Our emulation involves routers, packet  
capturing, monitoring, etc., and they also involve extensive  
background traffic generations.  They only stop short of testing in  
the real production networks -  which i can't since the testing  
requires overloading the production networks.  You might argue that  
our topology is not diverse enough, but well, i don't have funds/tech  
support to make it more diverse.

I am not sure whether those *IETF standard* TCP algorithms were  
tested as extensively as CUBIC/BIC. I doubt that the process taken at  
that time were as much rigorous as we are with CUBIC/BIC.  I am not  
sure either whether it is the job of  IETF to prove it is safe and  
harmless-- how do they know?  When those "standard" algorithms are  
IETF standardized, had they more evaluation than CUBIC/BIC? At best,  
they had ns-2 simulation. Back then there is no definition of  
realistic traffic patterns. I agree that our tests may not be enough  
to satisfy the IETF champions, but what is enough and how do they  
know without muddling their hands with testing?