Re: [ledbat] Note well on the BitTorrent problem on the example...

I know your example is about a single full-rate TCP flow, not  
BitTorrent.  But your all about the "bad queue" congestion problem,  
with occurs ALMOST exclusively for people in such situations when  
doing P2P activity over web-surfing.

I'd guess that part of the reason we have gotten away with such  
grossly asymmetric links is that, outside P2P usage, I'd guess that  
most users are grossly asymmetric in usage.  I know I was, the 3-s  
uplink buffer on my DSL line only bothered me when I was uploading  
photos or synchronizing work.

On Nov 22, 2008, at 7:41 PM, Stanislav Shalunov wrote:
>> it really should be "leech only",
>
> "Should"?  It clearly gets higher download rate for young swarms  
> with uploading.  Even in the most selfish interpretation it  
> shouldn't be leeching only.
>
>> as it contributes nothing to the P2P system in practice.
>
> Incorrect.  It contributes close to 128 kb/s, which is not nothing.

Its Amdahl's law-related:  It really IS close to nothing, and enough  
that if you care about user performance because you are experiencing  
head of line blocking issues, that reducing it to 0 is fine.

EG.  You have 1/2 the peers behind cable-modems, with 1 Mbps uplinks.   
And you have 1/2 the peers behind slow DSL links, with 128 kbps uplinks.

The P2P performance becomes limited by the uplinks in aggregate.  And  
the cable modems contribute 8x the aggregate bandwidth that the DSL  
links do.

In this case, putting ALL the DSL links to 0 upload will have a  
remarkably low effect on aggregate performance: a degradation of total  
system upload bandwidth of only 11%.

_______________________________________________
ledbat mailing list
ledbat@ietf.org
https://www.ietf.org/mailman/listinfo/ledbat