[e2e] Technology marches forward at the expense of the net?

[Michelle Munson]

"David P. Reed" wrote:
> 
> At 11:15 AM 12/14/2001 -0500, Tarik Alj wrote:
> >- Would it be possible to evaluate in what proportion, on the average, a
> >packet that gets through will contribute to the final file?
> 
> This isn't an FEC, exactly. There's an analogy here, though. The reason is
> that the recipient stops listening as soon as they have enough to
> reconstruct the file.  Every received packet contributes to the final file
> essentially 100%.
> 
> >- In other words how much bandwidth is wasted by applying FEC at such a high
> >level?
> 
>  From the recipient's point of view, essentially none.  However packets
> that get part of the way to the destination consume "bandwidth" (using that
> term is incorrect here, because we aren't talking about bandwidth, but
> bitrate).  So if few packets get all the way from source to destination,
> there can be heavy overhead before the bottleneck point or error sources,
> which increases as the source transmission rate gets much larger than the
> destination reception rate.
> 
> >- what about sequences of small file transfers, such as web pages?
> 
> This approach doesn't help at all with independent request of small files
> like web pages or transaction RPCs.  And hurts more as the web page size
> gets smaller, because if RTT*(sending rate) is large, the sender will send
> a lot of redundant packets after the recipient has received the whole web
> page, but it can't stop sending until it knows that the recipient has
> received the whole file.
> 

However better codes optimized for better performance on small files
(e.g. less redundancy overhead necessary to reconstruct the original) 
are always possible ... hint, hint :-)

Michelle Munson
An engineer at Digital Fountain