On Mar 10, 2009, at 10:16 AM, Arnaud Legout wrote:
Hi,
I believe that using caches (using regular clients on dedicated well
placed machine) makes sense to
reduce last miles traffic.
However, I do not see a scenario in which caches are useful (here I
mean that they cannot be replaced by
a real P2P client with an appropriate locality definition used by
the P2P protocol because the ISP architecture
prevent real P2P clients to be located at an appropriate location)
without requiring a large number of such caches.
In particular, deploying such caches can be much more expensive that
what you present.
If you take for instance the case of a DSL provider, you need to
place one dedicated cache per DSLAM.
As there is a lot of DSLAM for a single ISP (hundreds to thousands)
the cost is not negligible.
One may argue that even a few millions of dollars is not significant
for an ISP, but if it is to reduce a cost shift
from content providers to network providers, it might be hard to
convince ISPs that it is up to them to invest
in a new infrastructure.
Do you have in mind specific scenarios in which few caches can save
a lot
of last mile traffic, or did I misunderstood your proposition.
The caches don't need to be at the DSLAM, but can be "one up" from the
DSLAM/CMTS.
THe cache savings is not on the last mile download, but the last mile
UPLOAD which P2P requires, and it is this last-mile upload which make
P2P especially unattractive on cable networks.
The "Second to last mile" cost of bandwidth (from the central point
within the ISP to the DSLAM) is the only part that would be saved by
locating the caches at the DSLAM rather than slightly more centrally
in the ISP.
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
