Alan G. Nemeth
Corporate Consultant,UNIX Architecture and Technology

"The Internet is dying." I feel quite confident you will regularly see articles with this message in the industry and general press over the next few years. The message won’t be as new as the authors of the articles might believe, and the work to remove the most frequently identified problems was begun years ago within the Internet Engineering Task Force (IETF). Internet Protocol version 6 (IPv6) is a large family of protocols that form the basis of the IETF response to a set of problems identified in the early 1990s and for which the need is accelerated by the explosion of Internet usage.

One of the major concerns about the current Internet is the limited amount of address space. The underlying address for IP endpoints is 32 bits wide, permitting a total of 4 billion distinct addresses. Although this number seems large (and it seemed truly gigantic in the early 1970s when the width was selected), it is currently a real, practical barrier to current deployment patterns. Large users of Internet addresses can no longer get the address space they need for assignments. Because the Internet has run as a decentralized organization over the years, there is no effective central administration to support competition for scarce resources such as address space. Instead, the response of the community is to provide resources sufficient to keep allocation as a low-overhead activity. So IPv6 defines an address space of 128 bits. This currently seems like a gigantic number!

But limited address space is hard to build into a persuasive case for change. End users are much more likely to be concerned about the local problem of getting just "one more address," rather than the problems of keeping the Internet as a whole alive and functioning. So the IPv6 design deliberately incorporates a set of functionality improvements that provide attractive end-user capabilities. IPv6 includes much easier schemes for assigning addresses, which will reduce the administrative burden for users and their network managers. IPv6 provides a better framework for encryption and an expectation that it will be widely available and used. And IPv6 provides some systematic mechanisms for describing requests for specific quality levels in the service offered by the transport provider. These capabilities will address some very real, practical problems that do afflict individual end users of the Internet.

However, there is no expectation that it is acceptable to switch the set of Internet users to IPv6 either simultaneously or even over an extended time period. IPv6 must interoperate with the current installed IPv4 protocols for an indefinite period. This implies services that translate between the different addresses (and address assignment approaches that ease mechanical derivation of IPv6 addresses from IPv4 addresses), dual protocol stacks to permit communication with both protocols depending on the capabilities of the participants in the conversation, and schemes to accommodate security mechanisms and quality of service requests.

The entirety of IPv6 represents a large implementation effort to be undertaken by many different organizations. The Internet represents the largest example I know of a distributed computation that has survived for 27 years. (I date from 1969 when the first ARPANET [Advanced Research Projects Agency Network] nodes were installed.) With a few notable exceptions, this computation has run continually, despite constant changes in hardware, software, implementers, and operators. It has survived explosive growth far beyond the designs of its originators. It has done so with a volunteer organization driving the development direction. The community spirit has been crucial to making this work. IPv6 is an example of that community at work; no one organization can implement it all, either at a product level or at a deployment level.

The IPv6 paper in this issue describes the technical design needed to build an IPv6 implementation for the core protocols under the Digital UNIX operating system. Digital has been one of the leading prototype builders of the design specifications as they have evolved in the industry debates. At the time the Internet Protocol Next Generation (IPng) Directorate officially adopted key elements of the protocol, Digital’s implementation was the only one running to demonstrate that the design was indeed feasible. But we don’t believe that we can implement all the pieces of IPv6 as a single company. Therefore we choose to share the implementation experience through this paper to aid others in their efforts to deal with the implementation problems. We also don’t claim completeness; the full suite of specifications for IPv6 is evolving, and the software to implement it is large. We fully expect that portions of our ultimate product offerings will be developed by others in the industry.

The long-term evolution of the Internet captured in the IPv6 implementation paper is but one example in this issue of the extent to which computing now has a history that gives us much insight into the future. Certainly the paper by Supnik and Burnet is an explicit trip through computing history. The re-creation, both physical and logical, of computing systems of the past can only help remind us that the artifacts we create have a longer life than we anticipate. As our programmers write new code, or our hardware designers produce new architectural approaches, or our storage designers push the boundaries on new media technologies, do they consider the imponderables of running these systems 25 or more years in the future? The view of archivists trying to preserve this history reminds us of the difficulty of preservation after the fact and of the amazing duration of design decisions.

The paper on the evolution of Fortran is yet another example of the rich history of computing. Here we see clearly the evolution of a key language to accommodate the changing patterns of system architectural designs and parallel program concepts. The computer industry frequently develops commercially important programs by evolution --- the 100,000-line program that 10 years later has become 10 million lines of code in an assortment of languages and computing styles. Here the venerable Fortran (first introduced in 1954!) adds support for some of the latest approaches to fast system interconnect represented by MEMORY CHANNEL and the parallel architectures of clusters of SMP systems.

MEMORY CHANNEL reappears in the paper about TPC-C performance on TruCluster systems. This paper, one of a pair on the issues of tuning a commercially important benchmark, presents an attractive model for the benefits in performance that can be derived from a very fast interconnect and software structures to match. The performance levels achieved shatter world records on a benchmark that has had extensive attention and work.

The other paper on TPC-C performance with very large memory (VLM) illustrates the truth of an old design maxim, "If memory is getting cheaper, use more of it!" When Digital first built a 2-gigabyte (GB) memory board, it took more than a million dollars’ worth of DRAM chips to populate the initial instance. However, memory prices have continued to drop sharply, and today over 40 percent of the AlphaServer 8400 systems ship with 2 GB or more of memory. The memory prices will continue to come down, and the insights offered in this paper will help in understanding where additional memory can provide real benefits to customer workloads.

The final paper in the collection is on the AltaVista Forum approach to collaboration among groups exploiting the Internet and WWW technologies and brings us back around to the initial thoughts in this foreword. The ubiquitous nature of the Internet permits and encourages tools such as this that utilize computer systems in new ways. This approach builds on the fabric that we emphasized in the IPv6 paper but sees the Internet as a tool and a component of a larger solution and shows how to exploit these capabilities to allow new ways of working. Using imagination and building on the work of others are characteristic of the approach taken by those who are catalysts in the industry. The paper demonstrates how easy it is to build a system that would have been a major project just five years ago. This ease of construction is a benefit of the programming techniques and infrastructure investments and a spur to keep doing more of it.