Technology Review - Published By MIT
Advertisement

Bandwidth on Demand

An academic internet provides clues about ways to improve the commercial Internet.

By Erica Naone

Thursday, February 14, 2008

smaller text tool iconmedium text tool iconlarger text tool icon

Internet2, a nonprofit advanced networking consortium in the United States, is designing a new network intended to open up large amounts of dedicated bandwidth as needed. For example, a researcher wanting to test telesurgery technologies--for which a smooth, reliable Internet connection is essential--might use the network to temporarily create a dedicated path for the experiment. Called the dynamic circuit network, its immediate applications are academic, but its underlying technologies could one day filter into the commercial Internet, and it could be used, for example, to carry high-definition video to consumers.

Big sender: Internet2’s dynamic circuit network will help provide channels for large quantities of information to flow to and from academic research projects, such as CERN’s hadron collider, above. In the future, the technology may find commercial applications, such as for fast transfer of high-definition online video.
Credit: CERN
Multimedia
•  Download PDF

"The idea here is to basically look at the network in a different way," says Rick Summerhill, CTO of Internet2. The Internet Protocol (IP) currently used for the Web breaks data into packets that are sent through fiber-optic cables to their ultimate destination. The packets don't have to take a common path through the network; routers act like way stations along the network, examining every packet individually and deciding where it should be sent next. The problem with this system is that large data transfers can clog the routers with packets waiting for direction, and if the packets don't make it to their final destination at the same time, the receiver may experience jitter--interruptions to the data stream that can produce skips in online video, for example.

Summerhill says that, using the dynamic circuit network, a researcher could set up a temporary connection to another location that would work like a phone call: the user's data would be carried directly to that other location, uninterrupted by the traffic of others sharing the network. The result is that large quantities of information could be transferred quickly and clearly.

Story continues below

The dynamic circuit network is really an enhancement of a traditional network, rather than a replacement. Internet2 still has a backbone that uses the standard IP common across the Web. What makes the dynamic circuit network different is that it uses a circuit-switched network, which can be set up so that all the packets follow the same path. Also, those circuits don't have to be in place permanently. Lachlan Andrew, a researcher at Netlab, at the California Institute of Technology, explains that a circuit-switched network determines a pathway for the entire stream of packets, so that at every way station, they can be sent on without having to be individually examined. "Internet2 is developing technology to communicate between nodes, find a path, and construct it," he says.

The idea of the dynamic circuit network, Summerhill says, is that these circuits can be set up on demand, so that traffic needing excellent quality of service can step out of the regular flow. Because data is sent down fiber-optic cables at different frequencies of light, he explains, data from the dynamic circuit network can coexist with IP data and wouldn't require new cable to be laid. Summerhill says that Internet2 is working on software that could eventually be built into network devices to control these different flows and to set up circuits when and where they are needed.

Comments

  • Dynamic circuit switching is really old
    In the early 1970s, when the Internet (actually ARPAnet) was in its birth pangs, there were lots of debates as to which was the better way to go for packet switching. The debate was generally expressed as "datagrams" (lone packets, like IP) and "virtual circuits" (dynamic circuit switching).

    I'm surprised by the assertion that the big network providers are wary of dynamic circuit switching because they don't know how it will perform outside an academic environment. That was actually the rap on datagrams in the 1970s. The academic community was pushing datagrams for a variety of reasons (some technical and some not), but there was no experience with a heavily-loaded datagram network for about a decade after datagrams became the ARPAnet standard. The network providers favored dynamic virtual circuits; again, not all the reasons were technical, but the ability to control a heavily-loaded network was the major one.

    There are quite a few examples of dynamic circuit switching (think of it as pre-routed packet switching), both in history and the present. A couple of examples:

    (1) In 1976, about the time the Internet standards started, the CCITT (now the ITU, the standards body for the network providers) adopted its own packet switching standard, X.25, which used virtual circuits rather than datagrams. Several networks were implemented with X.25 before the Internet got big enough to dominate packet switching.

    (2) Asynchronous Transfer Mode (ATM) is a common way for Internet backbone providers to derive circuits between nodes. Those circuits are pre-routed packet circuits.

    Hope this gives a little background for the discussion.

    Cheers!
    DaveT
    Rate this comment: 12345

    dtutelman
    02/14/2008
    Posts:57
    Avg Rating:
    4/5
    • Re: Dynamic circuit switching is really old
      Thanks very much for adding the background. I should note that the people at Internet2 are aware that circuit switching is an old idea. Summerhill told me during our conversation: “What we’re doing is looking at ideas that were used in the past to see how we might enhance the existing network.”

      However, the dynamic circuit network itself, as built by Internet2, does seem to apply this idea in a new way, with new technology. For example, their control plane software, which sets up optical circuits across a variety of regional networks, is one of the things Summerhill told me is still in development.

      -- Erica Naone
      Rate this comment: 12345

      Erica Naone
      02/14/2008
      Posts:43
      Avg Rating:
      4/5
    • Re: Dynamic circuit switching is really old
      I am interested in your comments on dynamic circuit switching as I have been working in this area and view that these systems are the only ones that can scale to infinity.  I have been working on a layer 4 technology that will solve the problems of layer 3.  Please see my site:  www.fluidnetworking.net.
      Rate this comment: 12345

      samfwood
      02/16/2008
      Posts:1
      Avg Rating:
      1/5
      • Re: Dynamic circuit switching is really old
        I pretty much agree, based on studies my group did in the 1970s.

        If you're interested in the origins of Label-Switched Paths and hardware packet switches, A.G. (Sandy) Fraser invented Datakit at Bell Labs in the early 1970s. Different protocol and shorter packets and, of course, smaller and slower because of technology and lower demand. But all the fundamental concepts were there.

        Dave Tutelman
        Rate this comment: 12345

        dtutelman
        02/19/2008
        Posts:57
        Avg Rating:
        4/5
  • Enron?
    Didn't Enron have a similar business unit? They couldn't make it work, but it was probably not just the technology.
    Rate this comment: 12345

    atlacatl
    02/14/2008
    Posts:1
  • One size technology does not fit all
    The hype of the marketplace often overshadows technological common sense. Every time some new technology comes along, it's  hyped as the one solution to all problems, and it's NOT! Never!
    Twenty (and more) years ago, we had a digital switched phone system using time division multiplexing, a broadcast TV system, and a packetized protocol computer network.
    As LANs became more popular, Ethernet, Token Ring and other less popular networks were told they were obsolete, to be replaced by ATM, the backbone protocol of the digital phone system, to interface with dial up communications access. ATM killed off some very powerful networks like FDDI with false promises - then fell by the wayside when everyone realized a 53 byte ATM packet would have about 30% overhead, compared to a few percent for a 1000 byte Ethernet packet.
    Recently, we have had media wars - copper, fiber, wireless, with each saying it's the best, when what smart users are doing is combining the three to create networks that support users with different needs.
    I've never understood IPTV. Sure, it can be done, but it makes no sense to packetize a broadcast network. Hollywood is working on a better solution, a video backbones connecting studios to sound stages and eventually theaters. Keeping the video separate enhances quality and reduces overhead.
    Even VoIP causes problems for data networks, as latency causes problems, forcing higher bandwidth solutions at higher costs.
    In truth, we need several Internets, running on the same fiber using WDM (wavelength division multiplexing), data as IP traffic, switched voice and streaming video.
    The model of Verizon's FiOS fiber to the home using PON and WDM is a model for what can be done cheaply.
    I'll bet that optimizing traffic on different "carriers" saves more $ than trying to force all traffic on one backbone.
    Rate this comment: 12345

    fiberman
    02/14/2008
    Posts:73
    Avg Rating:
    3/5
    • Re: One size technology does not fit all
      You are absolutely right. My group did studies 29 years ago, comparing three types of networks (circuit switched -- then already in widespread use for five years, datagram, and virtual circuit) and three types of traffic (voice, interactive data, and bulk data/video). Two things we found were:

      (1) If you have a small amount of one kind of traffic added to a large amount of another kind, then one size does fit all. "Fit in" the small amount onto a network designed for the large amount, and let the small traffic reap the benefits of economy of scale.

      (2) If you have comparable amounts (same order of magnitude) of different kinds of traffic, none of the network types is best taken by itself. The best solution is a hybrid of multiple networks for different kinds of traffic.

      The study was done for WAN traffic, where transmission costs are a significant percentage of total costs. The answer may be different for LANs.

      Applying this to some of the history that you mention, VOIP was a wonderful, inexpensive, effective service as long as it was a tiny fraction of the bits flowing over the Internet. But, as it grows to a visible fraction, the traffic characteristics of voice are starting to affect the flow of traffic, and datagrams (IP) are not the best answer to a voice-only network.

      The study was done for WAN traffic. The answer may be different for LANs.

      Almost 30 years has flown under the bridge since then, and the technology has changed. The best choices for each kind of traffic could be different. But I'm sure that #1 and #2 above remain true.

      Cheers!
      DaveT
      Rate this comment: 12345

      dtutelman
      02/15/2008
      Posts:57
      Avg Rating:
      4/5

Log In

Forgot your password?     Register »
Advertisement

Videos

Making 3D Maps on the Move
Technology Review November/December 2009

Current Issue

Natural Gas Changes the Energy Map
The United States has vast supplies of this cleaner fossil fuel. But how should we use it?
Featured Content
Sponsored by:
White Papers

Twelve ways to reduce costs with SQL Server 2008
Find out how to reduce costs and get more efficient

Download

Total Economic Impact of SQL Server 2008 Upgrade
Forrester reports on increasing productivity and management capabilities

Download 

Achieving Cost and Resource Savings with UC
How Office Communications Server R2 and Exchange Server can make your business smarter and more efficient

Download 

The Compelling Case for Conferencing
Read how you can improve workload support and find IT efficiencies

Download

How Windows Server 2008 R2 Helps Optimize IT and Save you Money
Read how you can improve workload support and find IT efficiencies

Download

Windows Server 2008 R2 Hyper-V Live Migration
See how Windows Server 2008 R2 and Hyper-V enable virtualization and Live Migration

Download
Advertisement
Subscribe to Technology Review's daily e-mail update. Enter your e-mail address

TECHNOLOGY RESOURCES
Advertisement
MIT Massachusetts Institute of Technology © 2009 Technology Review. All Rights Reserved.