Next Generation Internet Technology

By Jimmy Kyriannis

The History of IPv6

In early 1992, the Internet Engineering Steering Group (IESG) documented that, at the then-current rate of IP address consumption, the Internet was at risk of depleting the number of available IP addresses and overburdening its national backbone network within just a few years.¹ In response, technologies such as NAT (Network Address Translation) and CIDR (Classless Inter-Domain Routing) soon emerged to dramatically alleviate the short-term concerns by enabling large numbers of computers to use just a few IP addresses, and by greatly reducing the amount of network information the Internet backbone routers² had to process. Coupled with improvements in router technology, the net result of those innovations was that the impending Internet failure predicted by the IESG was considerably deferred.

The long-term solution to supporting the growth of the Internet, however, lay in replacing the then- current communications system (Internet Protocol Version 4—IPv4), with a next-generation technology. Over the following two years, many competing protocols were developed and proposed to replace IPv4. Initially called IPng (next-generation), IPv6 (Internet Protocol Version 6)³ was ultimately recommended for adoption by the Internet community in early 1995.⁴

IPv6 offers a variety of enhanced features, including:

• significantly expanded addressing capacity (with roughly 3 x 1038 total addresses in IPv6 globally, compared with only 4.3 x 109 [4.3 billion] with IPv4)
• built-in support for authentication and privacy
• built-in support for mobile communications
• Quality of Service (QoS) prioritization of communications
• plug-and-play auto-configuration of network settings

Through a dual-stack mode of operation, the IPv6 standard requires that computers be able to use both IPv4 and IPv6 addresses simultaneously. This dramatically eases the transition from IPv4 to IPv6 technologies and, in fact, allows them to coexist on a computer and network without the need for additional system network interfaces or complex network routing manipulations.

In 1996, a year after the IPv6 recommendation was made, an experimental IPv6 network was developed on the Internet, called the 6bone, in which IPv6 was used for the first time to transmit data globally. This was achieved by embedding IPv6 communications within conventional IPv4 transmissions, through a technique called tunneling. In early 1997, New York University became one of the first U.S. universities to join the 6bone. Though highly experimental, this early work with IPv6 enabled NYU network engineering staff to participate in the technology development process and gain an early understanding of this new communications protocol.

Over time, however, faster and more reliable native (non-tunneled) IPv6 connections began developing between organizations, and native IPv6 access became available to research and educational institutions via the Internet2 (I2) network. Global use of the tunneled 6bone network declined as a result.

In early 2005, NYU leveraged our IPv4 connection to Internet2 via our local service provider, NYSERNet, to begin conveying global unicast⁵ IPv6 connectivity. With the availability of this native IPv6 connection, we began building IPv6 networks capable of supporting a greater number of features and considerable amounts of bandwidth through the use of commercial-grade network routers. Prior use of such devices as UNIX systems acting as routers was sufficient for proof-of-concept and informal or experimental use, but suffered from reliability, complexity, and performance issues.

IPv6 at NYU

Since our adoption of IPv6 connectivity over Internet2, IPv6-related activities at NYU have accelerated, particularly in support of research. One of the first IPv6-enabled research networks supports our supercomputer, Max. Unveiled in September 2005, Max is currently the fastest supercomputer in New York City, supporting various research projects with substantial computation requirements.⁶

With the installation of Max, NYU became the first university in the United States to provide global IPv6 connectivity to a supercomputer. We expect that this connectivity will foster international collaborations with organizations that already have IPv6-enabled networks, and in the future will facilitate access to IPv6-available data sets. Similarly, NYU's Information Technology Services (ITS) is now investigating how to make NYU Digital Library content accessible via IPv6 and IPv4.

As NYU has progressed with the construction of IPv6 networks, interest from our faculty has developed in kind. Faculty in the Interactive Telecommunications Program at the Tisch School of the Arts are interested in the use of IPv6 with wireless applications and in the possibilities that IPv6 end-to-end security⁷ might offer. Researchers in NYU's Computer Science department (CS) are studying IPv6 technology from a security and network performance perspective, particularly the benefits of very large IPv6 packets⁸ known as jumbograms⁹ in the development of real-time visualization technology.

NYU's Innovative Use of IPv6 Multicasting

Thus far, the above projects are making use of IPv6 unicast technology, although preparations have been made to support IPv6 multicast technology on a limited basis as well. As IPv6 multicast is still very much a technology in development, support for it by network hardware manufacturers is limited. End sites wishing to interconnect with others via IPv6 multicast often must do so with tunnels, as described earlier.

This past fall, however, NYU was established as the first end site in North America with native global IPv6 multicast connectivity. In collaboration with NYSERNet, NYU's IPv6 multicast connectivity to Internet2 and global research networks is accomplished though direct gigabit Ethernet communications without the use of tunnels. The end result is more reliable multicast transmissions with the ability to support multicasting at higher data rates. Streams of high-bandwidth Digital Video over IP (DVIP) to the global community, for example, can be more readily supported in IPv6 multicasts. We have since begun testing IPv6 multicast-capable applications such as Digital Video Transport System (DVTS) and VideoLAN Client (VLC) for video transmission, as well as the Microsoft ConferenceXP media conferencing application.

As a follow-up to this work, at the Internet Member's Conference in fall 2005, NYU and NYSERNet were jointly the first to successfully demonstrate the production use of native IPv6 multicasting. In this demonstration, an MPEG-2 video stream was multicast from NYU, a second from New York City by NYSERNet, and a third from Ann Arbor, Michigan by Internet2, while NYSERNet also multicast a 30Mbps DVTS video stream from Syracuse, NY. All four multicast streams were viewed simultaneously at the conference.

Through a similar collaboration with networking staff at the University of California (UC) Berkeley, NYU is looking to expand our tests of DVTS and other high-bandwidth technologies to include cross-country transmission of data via IPv6. We have already successfully demonstrated the IPv4-based transmission of four simultaneous digital video streams, operating at 120 Mbps, between the UC Berkeley campus and NYU over Internet2, in support of distance learning at the NYU Medical Center.

Additionally, a recent collaboration between ITS and the NYU's Television & Media Services TV Center resulted in NYU becoming the first to place an IPv6 multi- cast video channel on the Internet2 Multicast DV-GUIDE.¹⁰ Through the creation of this NYU channel, we are now able to work with interested members of the NYU academic community who would also like to list IPv6 and IPv4 multicast audio/ video content they wish to display globally on this Internet2 resource website. Through these and other activities, we hope to not only gain familiarity with IPv6 and advanced networking technologies, but also to facilitate academic endeavors in the process.

Future Goals and Challenges

Despite these successes, IPv6 technology continues to be very "bleeding edge," and is by no means straightforward to implement at a variety of levels. For instance, supporting jumbograms on a network brought on some new challenges, as many of our network devices were incapable of supporting packet sizes larger than the conventional 1500 bytes. This limitation would cause devices to misinterpret the jumbograms as network errors. Through some hardware reshuffling and upgrades, we have been able to position infrastructure to provide 9000-byte packet connectivity from CS research labs to the edge of NYU-NET. As 9000-byte Maximum Transmission Unit (MTU) framing is currently the highest common denominator supported by network manufacturers, an IPv6 jumbogram from a capable NYU network can now travel unimpeded to the Internet2 backbone network and reach other I2-connected universities or research organizations similarly capable of supporting jumbograms.

Another issue involves the propagation of IPv6 multicast packets on networks. Conventionally, multicast packets flood a network switch¹¹ unless the involved components are aware of multicast activities occurring on the network and take steps to prevent multicast communications from reaching unwanted recipients. As network switch manufacturers are only recently beginning to widely support IPv4 multicast, few, if any, are capable of doing so in an IPv6 multicast environment, making it difficult to offer IPv6 services on NYU-NET to a significant degree.

Finally and most notably, application support for IPv6 is extremely limited, and operating systems are only now beginning to offer IPv6 capabilities. There is a growing amount of software available that offers IPv6 capabilities; as much of this software is open source-based, however, there are sometimes tradeoffs between features and software reliability.

Macintosh OS X supports IPv6 by default, as do the various BSD-based UNIX operating systems (FreeBSD, NetBSD, etc.). Unfortunately, though, their support for IPv6 multicast is already rather dated, and incorporating the latest capabilities requires manual kernel modifications at the source code level. Though Linux supports IPv6 unicast and multicast, those features are not available by default and must be enabled manually. Windows XP does not officially support IPv6, although an experimental Microsoft implementation can be enabled manually. Its capabilities are also limited, and to date, Microsoft operating systems cannot make use of IPv6 for DNS operations.¹²

As we continue to deploy IPv6 connectivity on campus in support of academic and research activities and new services, we will also endeavor to collaborate with other organizations involved in this endeavor. Recently, the United States Government Accountability Office met with NYU to learn about our experiences with IPv6 in order to aid them in writing their report to Congress on IPv6 technology. Through such relationships, we hope to not only foster IPv6 technology globally, but also further it as an enabler of research and education.

For additional information about IPv6 technology, see www.ipv6.org. If you are interested in discussing the possibility of using IPv6 in support of your research or academic endeavors, please contact the ITS Faculty Technology Services Center at its.ftc@nyu.edu or 212-998-3044.

Footnotes

1. Initially documented in the Internet standards document RFC1380 (www.ietf.org/rfc/rfc1380.txt).
2. A router is a network device which transmits data from one network to another.
3. IPv5 was in fact developed, as an augmentation to IPv4 in support of streaming multimedia. The Internet Stream Protocol Version 2 (ST2) is documented in RFC1819 (www.ietf.org/rfc/rfc1819.txt).
4. RFC1752 (www.ietf.org/rfc/rfc1752.txt)
5. Unicast communications are the type that most of us are familiar with: a source computer sends data to a single destination computer. In multicast communications, the same source data is replicated to a group of simultaneous recipients. The latter is more efficient (yet more complex) for one-to-many transmissions.
6. See "Maximizing Research," and www.nyu.edu/its/pubs/connect/fall05/ackerman_supercomputer.html for more about Max.
7. With end-to-end security, data is encrypted at every point between the sending and receiving devices.
8. Packets are small units of IP information transmitted over a computer network. Historically, valid packets did not exceed 1500 bytes (characters) in length.
9. Jumbograms are very large IP packets of as much as 64KB (kilobytes) in size for IPv4 and 4GB (gigabytes) for IPv6. Designed for high-bandwidth applications requiring very high throughput, they minimize the overhead that would be generated by the processing of larger numbers of much smaller packets.
10. The I2 DV-GUIDE is accessible at http://db.arts.usf.edu/dvguide.
11. A switch is an Ethernet network device that provides wired connections to tens or hundreds of computers simultaneously. A network data jack in an office or residence hall room is connected to a switch, which in turn, connects to a router.
12. The Domain Name System is a worldwide database that translates Internet Domain Names into their IP addresses (e.g., www.nyu.edu translates into 128.122.108.74).

Author Biography