MAPPING -- Measuring And Practically Predicting INternet Growth
Measuring and predicting growth in Internet
addressing, routing complexity and energy
the past decade the Internet has moved from a well-hyped 'new thing' in
the 1990s to become a central part of most first-world societies. Our
increasing connectedness is enabling new modes of inter-personal,
world-wide communication for individuals, businesses and governments.
Yet the Internet faces a number of significant challenges.
The Internet's current “IP version 4” (IPv4) has almost
exhausted the supply of addresses with which to route packets to new
destinations [Hain05, Huston08], yet migration to an improved “IP version 6” (IPv6) addressing scheme is slow and poorly understood [Perset08].
Exhaustion of publicly routed IPv4 address space in the next few years
is expected trigger a new marketplace for trading small and large
blocks of existing IPv4 address space [Beck10],
as companies (and entire countries) strive to remain connected while
continuing to add internet-enabled devices. However, current models are
inadequate for predicting the likely value and costs of this new market.
There is also growing demand for internet-related electrical energy
world-wide – both in core infrastructure (service provider
routers, switches and wired and wireless links) and end-user devices
(such as PCs, laptops, home broadband gateways, 802.11-enabled access
points, internet-enabled PDAs and cellular/mobile phones). However,
current estimates of this end-user energy consumption are based on
unwieldy manual surveys.
Lack of address utilisation models
community lacks plausible models for the current use of (and future
growth in demand for) internet addresses by devices attached around the
Internet's “edge”. The use of IP addresses is neither
homogeneous nor centrally controlled. The Internet Assigned Numbers
Authority (IANA) periodically delegates large blocks of contiguous IP
addresses (identified by address prefixes) to five
geographically-arranged Regional Internet Registries (RIRs), who
further subdivide and delegate to Internet Service Providers (ISPs) in
each region. ISPs then allocate addresses (singly or in contiguous
blocks) to their own customers. Although the RIRs know which address
prefixes are allocated, we have no way to tell how many internet addresses are actually used at any point in time, or which are used intermittently. Some address prefixes are used sparsely, others are almost full [RFC3194].
IPv4 address space consists of allocated space (provided by RIRs to
ISPs), routed space (address prefixes actually advertised as reachable
by the inter-domain routing system), and occupied space (IP addresses
actually being utilised). Allocated and routed space can be ascertained
from RIR records and current inter-domain routing activity
respectively. However, to say an address is occupied requires a
measurement; for example it may
and independently verifiable tracking of estimated IP address
utilisation levels will significantly improve society's ability to
learn lessons applicable to the roll-out of IPv6 address allocation
policies, guide regulators in understanding the future market in usable
IPv4 address prefixes, and predict patterns of energy consumption
around the Internet's edges. IANA and the RIRs provide public
information on their overall prefix allocations. However, to date
individual ISPs have had little regulatory requirement to collect or
reveal the details of their actual address space utilisation (distinct
from their address space allocations), and equally limited economic
interest in doing so.
- respond to IP layer management query packets (“pings”);
- respond to a web request (SYN to port 80), either as a server would, or with an error;
observed participating in two-way packet exchange. (A one-way exchange
is not sufficient evidence, as the source address may be
Selective probing and monitoring to estimate current utilisation
is possible, but impractical, to establish utilisation by enumerating
and actively probing all possible IPv4 addresses in allocated or routed
space. For example, probing half the 2^32 IPv4 address space (roughly
2x10^9 addresses) at 500 probes per second would take around 50 days.
However, this sampling period precludes building a picture of how
individual IP addresses are used on a daily basis. In addition, probes
may not be answered due to firewalls, temporary end-host outages or
transient packet loss somewhere along the way.
Initial work by Heidmann et al has shown promising results by sampling only subsets of allocated IPv4 address space [Heid08].
Our goal is to develop and deploy improved techniques for intelligently
sampling subsets of the IPv4 address space in a way that captures the
essential utilisation patterns in hours or days rather than months.
Frequent repetition of such sampling enables us to build large- and
small-scale pictures of how utilisation changes over time.
Estimating the actual number of hosts
crucial challenge remains the estimation of how many hosts are
represented by a given public IP address. A single physical host may
have multiple IP addresses – perhaps for redundant connectivity
(multi-homing) or to support multiple virtual hosts each with their own
IP addresses. Conversely, routers at the boundaries of home and
enterprise networks may hide multiple hosts behind single IP address
using Network Address Translation (NAT). Indeed, one response to the
depletion of available IPv4 address space may be a period of
significantly increased NAT deployment. Such information is rarely
documented anywhere outside of organisation that owns the physical
hosts and NAT-enabled routers. We will synthesise new techniques for
estimating the degree to which public IP addresses are actually
NAT’d addresses, and how many hosts may be hiding behind them.
Our project will not attempt to estimate the number of hosts never connected to the Internet.
will improve current lower bounds on estimates of total energy
consumption relating to Internet-attached residential and office
computing equipment. Current studies, such as [BA07], are typically based on sales of PCs or numbers of broadband subscriptions (ADSL or cable). Studies such as [Ketal01] combined these with manual surveys and audits of the number of hours per day devices are used, such as [DOE99, Wetal06].
These figures vary significantly from country to country, and vary with
time as new services such as VoIP and VoD become common. Since such
surveys are major undertakings, they are not performed frequently; [BA07] used data from the 10-year old (1997) study [DOE99]. Energy use of office equipment has been studied in an Australian context by [LHH09],
again by means of sales data and surveys. A further limitation of
manual surveys, which will be overcome by this project, is that modern
computers have many low power modes. A user may be able to report
the number of hours per day for which a computer is powered on, but
will not know the fraction of time that the machine is in a lower power
Our project will improve such energy use estimates by providing a
tighter lower bound on the number of power-consuming devices turned on
and active at a given time, and enable other organisations to use our
new probing techniques at low overhead cost to themselves.
Bertoldi, B. Atanasiu,
Electricity Consumption and Efficiency Trends in the
Enlarged European Union, Technical report, Institute
for Environment and Sustainability, European Commission
Report EUR 22753 EN, 2007.
Beckman, Beware the black
market rising for IP addresses, InfoWorld, May 3rd
1999, Residential Energy Consumption Survey (RECS):
Household Energy Consumption and Expenditures 1997,
DOE/EIA-0632(97), November, 1999.
|| T. Hain,
A Pragmatic Report on IPv4 Address Space Consumption,
Internet Protocol Journal, Volume 8, Number 3, 2005.
|| J. Heidemann,
Y. Pradkin, R. Govindan, C. Papadopoulos, G.
Bartlett, J. Bannister, Census
and Survey of the Visible Internet (extended),
Technical Report ISI-TR-2008-649, USC/Information Sciences
Institute, February, 2008.
The Changing Foundation of the Internet: Confronting IPv4
Address Exhaustion, Internet Protocol Journal,
||K. Kawamoto, J.
G. Koomey, B. Nordman, R. E. Brown, M. A. Piette, M.
Ting, A. K. Meier, Electricity
used by office equipment and network equipment in the
US, Energy, Volume 27, Issue 3, Page 255, 2001.
|| M. Lane, A. Howard, S. Howard, The Energy
Inefficiency of Office Computing and Potential Emerging
Technology Solutions. Issues in Informing Science and
Information Technology, Volume 6, pp. 795-808, 2009.
||K. Perset, Internet
Address Space: Economic Considerations in the Management of
IPv4 and in the Deployment of IPv6, Directorate for
Science Technology and Industry, OECD Ministerial Meeting
on the Future of the Internet Economy, Seoul, Korea, 17-18
||A. Durand, C. Huitema,
Host-Density Ratio for Address Assignment Efficiency: An
Update on the H Ratio, RFC3194, Internet Engineering
Task Force, November 2001.
||C. A. Webber, J. A.
Roberson, M. C. McWhinney, R. E. Brown, M. J.
Pinckard, J. F. Busch, After-hours Power Status
of Office Equipment in the USA, Energy, 31, pp.