[Iccrg] Re: [tcpm] ECN feedback discussion

[Scheffenegger, Richard]

Hi John,

Assuming that the mechanisms of DCTCP get deployed, the threshold could be a small static number of packets in queue (K ~= 0.17*C*d [pps] and [s]); have you had a read of the analysis paper http://www.stanford.edu/~alizade/Site/DCTCP_files/dctcp_analysis-full.pdf
 
If / how the simple dctcp queue marking could be modified with queue delay (PIE, CoDel) instead of queue depth as signal is still open. But these two algorithms currently assume a low density marking paradigm...

Best regards, 



Richard Scheffenegger



> -----Original Message-----
> From: John Leslie [mailto:john at jlc.net]
> Sent: Montag, 12. November 2012 14:48
> To: Scheffenegger, Richard
> Cc: Ingemar Johansson S; tcpm at ietf.org; iccrg at cs.ucl.ac.uk
> Subject: Re: [Iccrg] Re: [tcpm] ECN feedback discussion
> 
> Scheffenegger, Richard <rs at netapp.com> wrote:
> >
> > Alternative to using ect1 by the sender (allowing high density ecn
> > marking), what about the sender continuing to use ect0, but high
> > density ecn marking by routers changing that from ect0 to ect1 andi
> > ect1 to ce (or ect0 directl to ce on very high load)?
> 
>    A truly interesting idea! Deployable...
> 
>    Of course, we'd need research to determine a good thresshold for
> marking ect0 to ect1.
> 
> --
> John Leslie <john at jlc.net>