North Dakota State University (NDSU) is one of eleven colleges and universities in the North Dakota University System. With an enrollment of 10,000 students, NDSU is a land-grant institution offering several Phd programs and home to several internationally recognized researchers in the fields of chemistry, engineering, and agriculture. NDSU is a member of Internet2 and a partner in the Great Plains Network, a NSF-funded regional network connecting six mid-western states and supporting earth system sciences research. The authors are employed by the NDSU campus computer center, Information Technology Services, (ITS). Dr. Wettstein is the senior system administrator and Mr. Grosen is the associate director for networking and multi-user computing systems. ITS provides desktop computing support and services, training, networking, and other information technology to the NDSU campus. ITS is also home to the SENDIT network, a K-12 technolgy services initiative for the state of North Dakota.
As a rural state with a small population, North Dakota has faced and continues to face many challenges in deploying information technology services. In the mid 1970's, university system administrators conceived of the notion of tackling some of these problems by centralizing as many services as possible and deploying a communications infrastructure to provide access. The resulting system of services was named the Higher Education Computing Network (HECN). Initially, the HECN concentrated on providing administrative computing services and the University of North Dakota was designated the administrative computing host site. In the late 70's it was recognized that technology was becoming an important component in academics and, in 1980, NDSU was charged with the responsibility of centrally providing academic computing services. Initially, the infrastructure consisted of a single IBM mainframe computer. Over time, it came to include multiple servers and services such as email, domain name services, USENET news, Unix shell access, etc. When the mainframe system was removed from service four years ago, the model for delivering services remained essentially the same; users accessed services via interactive logins over the state's wide area networks.
As desktop PCs became more ubiquitous and the advent of graphical interfaces made these applications more accessible to the non-technical, users began to prefer using their desktop software over that available on the centralized servers. Concurrently, a survey of HECN users and administrators revealed that there was consensus on the need for a unified electronic ``phonebook'' or directory. The survey also revealed the desire for a standardized scheme for email addresses. Two criteria were identified: first, the recipient portion of the email address should be based, as closely as possible, on the person's full name and, second, the dependency on hostnames for mail servers should be replaced with DNS domain names and MX records. These observations and requests became input into a major re-design of the infrastructure for providing HECN academic computing services.
During the design process, additional technical requirements were developed. Key among these was moving to a client-server architecture using micro-servers running Linux. The old server infrastructure was based on several Unix hardware and operating system combinations. This requirement would allow us to reduce the heterogeneity in the server infrastructure, simplify management, ease development, and lower costs.
Another technical requirement was to move to a Category of Service (CoS) model. In the ``old'' model, users were assigned Unix logins which gave them access to all services (email, printing, etc.) supported on the server. Servers were, in effect, equivalent to certain standard sets of services and users were assigned according to their needs. In some cases users were assigned to multiple servers to provide them with all the services they required. In the client-server model, Unix logins could no longer be used. Thus, a new way of creating and assigning sets of services for users was required.
The CoS model was the answer to this problem. Services would no longer be tied to any particular server. Each user would be assigned a single ``login'' with a single password. Each service would be separate from all other services and enabled or disabled without affecting any other services for that user. Custom sets of services could be created for users without the need for multiple server logins.
Two fundamental problems arose out of these requirements: