Internet Traffic Flow Study Underway
January 9, 1998
By Jon Bashor, jbashor@lbl.gov
Funded by the National Science Foundation, the year-long pilot study of the
Internet structure and its performance could serve both to troubleshoot
problems and to eventually give users a means to rate Internet services and
providers. The projected is led by Vern Paxson of the Lab's Computing Sciences
Network Research Group and by scientists at the Pittsburgh Supercomputing
Center.
Researchers will place computers at various locations on the Internet to
automatically measure network performance between stations. This effort is
intended to provide the foundation for a much larger study of the Internet
infrastructure.
Berkeley Lab alone participates in about 500,000 connections every day, as
scientists rely increasingly on computer networks for scientific collaboration.
For example, Internet traffic has grown from a mere 45 web connections in
October 1992 to more than ten million last month, Paxson notes.
One of the biggest problems with the Internet today is congestion. The net's
worst problem, however, also reflects one of its main strengths. The myriad of
data connections in the network all share the available capacity among them,
rather than reserving capacity individually, as do telephone connections. This
sharing makes for a much more efficient use of the network, but it also leads
to traffic overload.
The data for a particular connection travels along a series of links, each step
being termed a "hop," with twenty or more hops in an end-to-end chain. "The
network is very good at hiding the individual hops a message takes from the end
users," Paxson says. "Links go down all the time and packets have to be
re-routed, which happens automatically."
Within the Internet research community, the web is known as a "success
disaster"-- something that was so useful that it grew faster than the system
could accommodate. The underlying problem is that the web, a huge hypertext
database, is poorly designed to run on the global Internet, Paxson says. As the
popularity of the web grows, the problems loom ever larger.
"We can build systems with hundreds of things and they work fine, but when we
scale them up to hundreds of thousands, they break down, and it's hard to
predict where they will fail." he says. "Users may have the perception that
something is overloaded when it's actually misengineered."
Without measurement, it can be impossible to tell where the problem lies. One
of the basic measurements provided by Paxson and his Pittsburgh colleagues is
the rate at which data packets are discarded, or "dropped" by the network,
something which directly affects performance.
Information on the net is sent via IP--the Internet Protocol. When too many
data packets come in, routers hold them in buffers until traffic thins out.
When traffic is too heavy, the buffers fill up and the router drops packets.
The IP provides the basic building block, which says, "You hand me a packet and
I'll do my best to deliver it," Paxson explains. "But no promises--it may
arrive as expected, it may get duplicated along the way, it may be corrupted or
it may arrive out of order. Although that sounds like a weakness, it's actually
a strength. It's very cheap to provide connectivity because you promise very
little."
On top of the IP layer is the Transmission Control Protocol, or TCP. This is
the mechanism that ensures that messages do in fact get there.
"On the web, when you do a data transfer and you get the desired item, the
transfer is finished," Paxson says. "When you're using a browser like Netscape
or Explorer and it stops, that means a packet was dropped. But TCP isn't giving
up, it's really backing off because there's a problem. When you get a message
back saying something can't be found, your packets did get through, but there's
a problem with another part of the system."
Paxson ultimate goal is to give Internet users a method whereby they could
click on a button and find out the root of the problem, and in the process
contribute to the measurement database. While most users know about the typical
problems, Paxson expects to find the unexpected.
In 1985, ARPANET (the forerunner of the Internet), was collapsing due to
congestion of data transmission. Berkeley Lab computer scientists developed
several Transmission Control Protocol algorithms to prevent them. The
algorithms became mandatory standards in 1989.
In 1986, the network was plagued with routing instability between the many
connecting systems, resulting in many transmissions being lost. Lab experts
developed a tool to trace data packets along the way, allowing routing problems
to be pinpointed and corrected.
As the Internet and the Web gained wider exposure, users began to look for
wider applications. The idea of using the Net to transmit both audio and video
was dismissed by many as impossible. The Network Research Group at Berkeley Lab
saw it as a challenge and helped created Mbone, a multicast virtual network
using the Internet. Today, more than 500,000 people around the world routinely
use the technology to conduct meetings, seminars and workshops. Protocols
developed here at the Lab are now Internet standards.
For more information about the Lab's Network Research Group, visit its website
at: http://www-nrg.ee.lbl.gov/.
For all the Internet users who wonder why e-mail sometimes bogs
down or why a favorite website takes an eternity to load up, a new effort by
computer scientists in Berkeley and Pittsburgh may start turning up some
answers.
Lab helps build a better Internet
For more than a decade, computer scientists at Berkeley Lab have analyzed
problems with the Internet, then come up with technological solutions to fix
them. In fact, many of the technical standards now used to keep the Internet
working smoothly were invented here at the Lab.
