RE: [dccp] The TFRC implementation in ns-2.28: variable size packets support

["Arne Lie" <arne.lie at item.ntnu.no>]

Dado,

Thanks so much for replying so fast, and for the references! I suppose the
best "work around" is to use small packets, so that the "dummy bits" for the
small P- and B- frames, and the last packet of a larger I-frame, are as
small as possible. But then again, small packet fragmentation generates even
more overhead, so...  the optimum solution as far as total overhead is not
easily calculated. Currently, I find this as a serious drawback, but I
suppose it simplifies the TFRC equation significantly.

Arne

> -----Original Message-----
> From: Dado Colussi [mailto:gdc at iki.fi] 
> Sent: 1. februar 2006 16:23
> To: Arne Lie
> Cc: dccp at ietf.org
> Subject: Re: [dccp] The TFRC implementation in ns-2.28: 
> variable size packets support
> 
> Arne Lie wrote:
> > Can anybody confirm the following finding in ns-2: the send 
> queue in 
> > tfrc.cc consist only on *number* of packets, assuming ALL 
> packets have 
> > equal size = "size_"!!??
> 
> 
> IIRC, your finding is correct.
> 
> 
> > If this is true, does anybody have any comments on this 
> issue before I 
> > implement a queue also holding packet sizes? E.g. should  I 
> not do it, 
> > because DCCP/TFRC is not supposed to support this?
> 
> 
> That modification alone is not likely to bring you far. You 
> should also modify the loss fraction computation and 
> currently there is no consensus
>   of how that should be implemented. Check out the CCR paper 
> of Widmer et al. [1] on the matter.
> 
> Cheers,
> Dado
> 
> [1] http://icapeople.epfl.ch/widmer/tfmcc/vp-tfmcc.html