From: "Bill St.Arnaud" <bill.st.arnaud@xxxxxxxxxx>
Date: February 20, 2006 6:18:30 AM PST
To: <news@xxxxxxxxxx>
Subject: [CAnet - news] FFTh study refutes some previous notions of
network utilization and symmetry
For more information on this item please visit the CANARIE CA*net 4
Optical
Internet program web site at http://www.canarie.ca/canet4/library/
list.html
-------------------------------------------
[Interesting study of actual users on FTTh networks. Note- in Japan I
believe that many VDSL connections are considered FTTh. The number
of true
FTTh connections is a lot smaller. Thanks to Dirk van der Woude,
Jaap van
Till and Andrew Odlyzko for this pointer -- BSA]
<http://www.iijlab.net/~kjc/tmp/rbb-20060211.pdf>
In Japan now some 4.5 to 5 million FttH connections
are in place, so research into the resulting traffic
patterns is quite interesting for (yet) fiber challenged
countries. The attached report to my knowledge is
the first in which the effects of wide spread FttH (i.e.
super high symmetric broadband) is measured.
Classic thought says that a small group of very heavy
users is responsible for the over large majority of traffic.
Hence the arguments of owners of copper and coax
networks:
"Fiber is not necessary, the over large majority of users
does not need it" and
"nobody asks for symmetric bandwidth".
The attached research clearly seems to point in another
direction: users that do not have high symmetric bandwidth
will not or less use bandwidth hungry applications.
The moment these network constraints are solved they
will at times become heavy users. Those 'heavy-hitters'
are (at least in Japan) not a small and clear defined
group - but anybody with access to symmetric super
high bandwidth.
So, quoting my Alderman Mark van der Horst who believes
in short statements, "one does not buy a Ferrari in a country
where roads are sandy paths".
greetings,
Dirk
===============
Our per-customer measurements reveal the behavior
of residential traffic in depth. At first, we noticed a
large skew in traffic usage: the top 4% of heavy-hitters
account for 75% and 60% of inbound and outbound traffic,
respectively. Fiber traffic accounts for 86% and 75%
of inbound and outbound traffic, respectively. We tend
to attribute the skews to the divide between a handful
of heavy-hitters and the rest of the users. Our in-depth
analysis, however, shows the existence of diverse and
widespread heavy-hitters who appear to be casual users
rather than more dedicated users. In addition, the total
traffic behavior seems to reflect the balance of the
diversity.
For example, the large skew in per-customer traffic
seems to be caused by a small number of heavy-hitters
but, in fact, the distribution of per-customer traffic follows
a power law and it is difficult to draw a line between
heavy-hitters and the rest of the users. The large skew
in traffic volume between fiber and DSL is not caused
by qualitative differences in the behaviors of fiber and
DSL customers but simply by the larger percentage of
heavy-hitters among fiber users. The large skew of userto-
user traffic in residential traffic seemingly points to
peer-to-peer file-sharing but it is apparently a mixture
of file-sharing and content-downloading. All the results
indicate that the perceived divides are actually caused
by diversity. At the same time, the entire behavior reflects
the balance of this diversity, but it is sometimes
dictated by the most influential group.
We can no longer view heavy-hitters as exceptional
extremes since there are too many of them, and they
are statistically distributed over a wide range. It is more
natural to think they are casual users who start playing
with new applications such as video-downloading
and peer-to-peer file-sharing, become heavy-hitters, and
eventually shift from DSL to fiber. Or, sometimes users
subscribe to fiber first, and then, look for applications to
use the abundant bandwidth. The implication is that, if
a new attractive application emerges, a drastic change
could occur in traffic usage. For example, current peerto-
peer applications do not take locality into consideration,
but future applications could as suggested in [13].
As for the generality of our measurements, several aspects
are specific to Japanese traffic. One is the high
penetration of fiber access. It seems to take some time
for other countries to deploy fiber access; even Korea
that has the highest broadband penetration ratio does
not have widespread fiber access [16]. Japan is a model
of widespread symmetric residential broadband access.
Another is fairly closed domestic traffic. The current
situation is partly due to language and cultural barriers
and partly due to rich connectivity within the
country. The former could be common to other non-
English speaking countries to some extent, and the latter
can be seen simply as the geographic concentration
of bandwidth-rich users.
7 Conclusion
The widespread deployment of residential broadband
access has tremendous implications on our lives. Although
its effects on the Internet infrastructure are difficult
to predict, it is essential for researchers and industry
to prepare to accommodate innovations brought
by empowered end users. Extensive effort to establish
protected data sharing mechanisms with commercial
Japanese Internet backbone providers has allowed
us to achieve an unprecedented empirical analysis of a
significant segment of the Japanese residential broadband
traffic.
The growth of residential broadband traffic has already
contributed to a significant increase in commercial
backbone traffic. In our study, residential broadband
traffic accounts for two thirds of the ISP backbone
traffic, which will force significant reevaluation of
the pricing and cost structures of the ISP industry.
We have further studied residential per-customer traffic
in one of the ISPs, and investigated differences between
DSL and fiber users, heavy-hitters and normal
users, and in geographic traffic matrices. We found that
a small segment of users dictates the overall behavior;
4% of heavy-hitters account for 75% of the inbound volume.
The fiber users account for 86% of the inbound
volume. About 62% of the residential traffic volume is
user-to-user traffic that exhibits impressively diverse behaviors.
The distribution of heavy-hitters follows power
law without a clear boundary between heavy-hitters and
normal users.
For future work, we will continue collecting aggregated
traffic logs from participating ISPs. We are also
planning to do per-customer traffic analysis from other
ISPs, and hope to compare our results with measurements
from non-Japanese ISPs.