Position Statement for
NSF Workshop on Internet Statistics Measurement and Analysis
Somewhere between the current practice in telephone networks and
the current practice
on the Internet backbone,
there is a place that provides sufficient collection of
information to adequately provision and understand the
demands on the Internet
backbone without much impact on the cost of the service.
Billing and accounting information collection and processing
is estimated to require an investment in hardware and
service that equals the investments needed to actually
provide the service for standard telephone service that
is billed by time and distance.
Clearly this is not a direction that is either feasible nor desirable
for the Internet backbone.
The small average transfers and the complexity of trying to
capture the usage to different locations and the highly
variable speed of transfers make it very unlikely that
we will do substantial useage and distance based billing any time soon.
On the other hand, current Internet backbone practice is to
charge a fee based on the speed of the datalink connection,
whether or not that connection is highly or lightly used.
This tends to lead to the situation where bandwidth is parsimoniously doled
out to the user because responsiveness is not necessarily part of the
equation and the provider is provided incentives to keep
the pipe as constrained as possible.
Aside from collecting information for billing and accounting
purposes, two other reasons to be concerned with data
collection on the Internet are
- to have information about network loading over
time so that provisioning decisions can be
made and implemented in a timely manner,
- to have data to guide equipment makers in the
decisions they make about what kinds of service and support are
needed for the ever changing applications that use the network.
The measurements needed for provisioning decisions are clearly something that
equipment makers must provide. In general this is aggregate
usage information. The
collected information may not be
made publicly available because the ability of a
service provider to use this information to provide
acceptable levels of service
without over investing in capacity
is one of the key elements of competition.
The collection of information that is useful to
the research and design community is another matter.
This information often needs only to be collected for short
intervals at widely different times in order to be useful. The
long term sampling can identify trends and performance requirements
that change slowly over time. This
collected data may require substantial off-line
analysis before conclusions can be drawn,
and different organizations may arrive at different conclusions
from the same data. Since this data can be collected
at only a single place, or at most only a few places,
in the network, it is not unreasonable to have some specialized
equipment for this kind of data collection.