Summary: The impact and implications of Growth in Residential User to User traffic

From: Kiran Kumar Gollu <kkgollu_at_cs.toronto.edu>
Date: Tue, 21 Nov 2006 02:21:09 -0500

The paper presents results of long-term traffic measurements collected
over 21 months from seven ISPs in Japan. The goal of the paper is to
obtain a better view of ratio of residential broadband traffic to other
traffic, changes in traffic patters and regional differences across ISPs.
It is important to note that the paper focuses on user-to-user traffic
rather than peer-to-peer file sharing.

The paper presents analysis of aggregated traffic as well as per-customer
traffic patterns. To analyze the aggregated traffic characteristics, it
uses data set collected by aggregating interface counters of edge routers
from seven ISPs. To analyze per-customer traffic, it uses sampled data
from one of the ISPs.

The aggregate traffic analysis results signify that RBB customer traffic
has increased by 37% from previous year. The outbound traffic to customers
is slightly larger than inbound, even though it is assumed that that home
users downstream traffic is much larger than upstream. Surprisingly,
evening traffic for residential users is higher than traditional office
customer traffic. Results for external aggregated traffic imply that
inbound traffic is higher than outbound traffic but this may not
applicable to other countries (due to external factors such as cultural
differences). Finally, prefectural (regional) traffic patterns exhibit
similar temporal behavior and traffic is roughly proportional to the
population the prefecture. More importantly, results also imply that there
is no clear regional concentration of heavy hitters.

The authors then analyze the per customer traffic. They found that number
of active DSL users is slightly higher than fiber users but there are more
heavy hitters among fiber users. The distribution of heavy hitters is
heavy tailed but distinction between heavy-hitters and normal users is not
clear. For e.g. top 4% of the heavy-hitters use 75% of the inbound traffic
and 60% of the outbound. This means that small group of heavy hitters
represent a significant part of total traffic. Also, distributions look
similar across different prefectures except for some slight differences
due to difference in the number of users. A thorough investigation of
differences between DSL and fiber users revealed that inbound traffic of
fiber heavy-hitters is much larger than outbound traffic. Also, fiber
traffic has lot of fluctuations. However, inbound traffic of DSL heavy
hitters is heavily saturated.

Few interesting conclusions:
1) High volume of traffic is not only generated peer-to-peer file
sharing systems but also by other means such as content downloading from a
single server.
2) Internet traffic has very poor locality, in contrast to telephone
communications where users talk to nearby near by neighbors.
3) Slight change in the algorithms or charging policies of ISPs could
have significant impact on total current traffic volume is heavily
impacted by extreme heavy-hitters. This forces significant re-evolution of
pricing and cost structures in ISP industry.

Finally, I thought some of results might be biased to Japan because some
technologies such as FTTH are available only in very few countries.
However, the paper brings out an important problem of managing the ISP
backbone traffic.
Received on Tue Nov 21 2006 - 02:21:58 EST

This archive was generated by hypermail 2.2.0 : Tue Nov 21 2006 - 02:23:11 EST