Re: [p2prg] New version of p2prg-mythbustering

Enrico Marocco <enrico.marocco@telecomitalia.it> Tue, 14 July 2009 10:14 UTC

Message-ID: <4A5C520C.6070502@telecomitalia.it>
Date: Tue, 14 Jul 2009 11:38:20 +0200
From: Enrico Marocco <enrico.marocco@telecomitalia.it>
User-Agent: Mozilla-Thunderbird 2.0.0.22 (X11/20090701)
MIME-Version: 1.0
To: "Ray.Bellis@nominet.org.uk" <Ray.Bellis@nominet.org.uk>
References: <OF0B76E4E5.7F557F12-ON802575F2.0057FF67-802575F2.00584D44@nominet.org.uk> <4A5B604B.5030006@telecomitalia.it> <OFAD1D99E6.3BAD6079-ON802575F2.005CD14B-802575F2.005CD7CE@nominet.org.uk>
In-Reply-To: <OFAD1D99E6.3BAD6079-ON802575F2.005CD14B-802575F2.005CD7CE@nominet.org.uk>
Content-Type: multipart/signed; protocol="application/x-pkcs7-signature"; micalg="sha1"; boundary="------------ms060000060609070401070002"
Cc: "p2prg@irtf.org" <p2prg@irtf.org>
Subject: Re: [p2prg] New version of p2prg-mythbustering
Precedence: list

A little slip notwithstanding -- (2 * 5.5) + (2 * 2 * 5) = 31 ;-) -- the
proportion sounds reasonable, but the two cases are fundamentally
different. In fact, while in the former the two on-net peers are
literally leeches, in the latter they upload at least half of the file.
With the same share ratio (0.5, that is still half the average), the
cost without localization would sum up to:

    2 clients * (backhaul down + transit)
  + 2 clients * 0.5 (backhaul up + transit)
  = 2 * (10 + 1) + 2 * 0.5 * (10 + 1)
  = 33 units

This makes sense, as in the case with localization (cost: 31 units) you
have spent the same amount for the backhaul, but you have paid transit
for 1 file only (the 2 halves the peers downloaded from the off-net
seeder) instead of 3 (the 2 entire downloads plus the two half uploads).

The bottom line here is that if p2p traffic localization does not
increase uplink bandwidth usage (this could be argued, we took a first
stab in section 5 and 5.2 but more thought is definitely needed), the
savings are proportional to the impact of transit on the overall per-bit
cost.

Enrico

Ray.Bellis@nominet.org.uk wrote:
>> Well, no, if the content was locally available (i.e. one seeder
>> existed), but the local leecher downloaded it from outside, the total
>> cost of the traffic generated by the two peers would amount to £220 as
>> the seeder would just be serving some other off-net leecher (and thus
>> costing another £110).
> 
> I was thinking more of a trivial example of two leechers on-net, and one
> seeder off-net.
> 
> Using a 10:1 cost ratio between backhaul and transit, and assuming a
> cost of "1 unit" for the cost of transferring that data over the transit.
> 
> If both leechers download everything from the seeder:
> 
>   = 2 clients * (backhaul + transit)
>   = 2 * (10 + 1)
>   = 22 units
> 
> If each leacher downloads exactly half of the data from the seeder, and
> the remaining half from the other leacher (assuming non-overlapping
> fragments):
> 
>   =   2 clients * 0.5 * (backhaul + transit)
>     + 2 clients * (up+down) + 0.5 * backhaul      [*]
>   =  (2 * 5.5) + (2 * 2 * 5)
>   =  33 units
> 
> so it appears that the on-net leeching costs 50% more than two downloads
> from off-net.
> 
> Ray
> 
> [*] I think I've done this right - each leecher downloads half of the
> file from the other leecher, but the ISP has to pay for it in both
> directions.

Attachment: smime.p7s

[p2prg] New version of p2prg-mythbustering Enrico Marocco
Re: [p2prg] New version of p2prg-mythbustering Ray.Bellis
Re: [p2prg] New version of p2prg-mythbustering Enrico Marocco
Re: [p2prg] New version of p2prg-mythbustering Ray.Bellis
Re: [p2prg] New version of p2prg-mythbustering Matthew Kaufman
Re: [p2prg] New version of p2prg-mythbustering Thorsten Strufe
Re: [p2prg] New version of p2prg-mythbustering Enrico Marocco
Re: [p2prg] New version of p2prg-mythbustering Arnaud Legout

Re: [p2prg] New version of p2prg-mythbustering

Attachment: smime.p7s