Stevens.Le_Blond at sophia.inria.fr wrote:
> In Section 3.1:
>
> You should distinguish between "local" cross-domain traffic (a single ISP
> and a single torrent) and "global" cross-domain traffic (all ISPs and
> 214,443 torrents). This is very important because locality may allow to
> save a lot of "local" cross-domain traffic in given scenarios without
> being beneficial overall. In other words, if it's not possible to save
> "global" cross-domain traffic, locality can be *mainly* useless.
>
> "... Simulations and field tests have shown a reduction varying from 20%
> to 80%."
>
> If you consider local cross-domain traffic, our experiments with real
> distributions of peers per AS actually show that you can save more than
> 99.5% of "local" Cross-domain traffic and 40% of "global" cross-domain
> traffic ([LeBlond], Section 6.3).
The distinction between observed "local" and "global" cross-domain
traffic reduction is actually fine and definitely worth pointed out. The
section about application performance is already explicit about the fact
that the results of the localization approaches are going to be strictly
dependent on the local network environment, so, besides clarifying the
nature of the numbers in your work, it may be a good idea to explicitly
highlight the expected differences in this section as well. Do you think
you can suggest some text and/or pointers for the Discussion section
about how the effectiveness of the locality strategy relates to the
network conditions? I guess that at the end of the day everything boils
down to the number of peers per AS/ISP, but maybe you've found other
interesting factors that impact too.
> "4. Simulations of the localization approach proposed in [LeBlond]
> run on data collected from crawling of real BitTorrent swarms
> indicate a 40% reduction of cross-domain traffic."
>
> We didn't run simulations but experiments with real BitTorrent clients.
Right, we fixed it somewhere else in the document after Aranud's
comments but did miss this one.
> The other works that you cite consider "local" cross-domain traffic so you
> shouldn't compare those results with 40% which is for "global"
> cross-domain traffic. I would suggest that you either create a subsection
> for "global" cross-domain traffic or make this difference very clear. In
> any case, you can't compare those numbers directly as they mean something
> different.
How about replacing the above with the following text?
4. Experiments with real BitTorrent clients run by researchers a INRIA
[LeBlond] have shown a 40% reduction in cross-domain traffic on a
global scale, with local peaks up to the 99.5% in exceptionally
favorable conditions.
> In Section 3.2:
>
> We have also shown that in the case of severe bottlenecks, BitTorrent with
> locality can be more than 200% faster than regular BitTorrent ([LeBlond],
> Section 5.3).
Ok, I'll add the reference there too.
> In case you didn't have a look at the second version of [LeBlond], here's
> the refence:
>
> Stevens Le Blond, Arnaud Legout, Walid Dabbous.
> Pushing BitTorrent Locality to the Limit.
> Technical Report (inria-00343822, version 2 - 12 May 2009), INRIA,
> Sophia Antipolis, May 2009.
> http://hal.inria.fr/inria-00343822/en/
>
> Hope this helps.
Sure it does!
--
Ciao,
Enrico
Attachment:
smime.p7s
Description: S/MIME Cryptographic Signature