[e2e] end of interest

rick jones perfgeek at mac.com
Sat Apr 19 09:51:50 PDT 2008

On Apr 18, 2008, at 1:47 PM, David P. Reed wrote:

> Andrew -
> I do exploration in Linux myself (I only run Windows in KVM, and  
> only when I absolutely have to).  Regarding the Major Company's  
> proprietary TCP code - I agree that minor tweaks to the TCP standard  
> are adopted as long as the Major Company's programmers are the ones  
> who act as gatekeeprs.  However, that's like what AT&T Bell Labs  
> used to say when they disparaged the Internet Experiment in the  
> 1980's - "when those lunatics working with Vint Cerf and Bob Kahn do  
> something really interesting, we'll put it in our network".

Speaking as an employee, but not on behalf, of a Major Company with  
proprietary TCP code, I think that is exactly the way a Major Company  
does behave.  And I don't see that as a Bad Thing (tm).  While some of  
our customers might want the latest and greatest, by far most of them  
want stuff that Just Works and don't care to be experimental  
subjects.  That means they don't want new stuff thrust upon them  
unbidden.  The developers with Major Co may still put some new/ 
experimental things in the proprietary TCP code, but the bar will be  
quite high.

"The network" (stack, intranet, whatnot) is much more critical/ 
valuable to Major Co's customers these days then say in '88/'89 when I  
and my colleagues were able to drop VJ Congestion control into MPE/XL  
TCP "just because it is better."  It was also quite easy to  
demonstrate just how much better things were - and as part of that we  
(the developers at Major Co) were able to look at the research "those  
lunatics" :)  published without using our customers as experimental  

The key today (and perhaps Catch-22) is to get the customers to  
request the features.

> If you have no hope of deploying most innovations without bargaining  
> with Major Co., then why bother doing research?

To demonstrate the the customers and employees of Major Co that the  
innovations are worthwhile.

An irony in all this is that the customers of Major Co end-up  
demanding, and getting, changes in the TCP stacks that would probably  
leave some researchers aghast.  Stuff like CKO, ACK avoidance  
heuristics, very configurable retransmission timers, etc...  And  
miracle of miracle their networks are not known to have imploded :)

rick jones
there is no rest for the wicked, yet the virtuous have no pillows

More information about the end2end-interest mailing list