By the end of this decade most major universities will have a wide variety of on-line multimedia learning resources, digital libraries will be well underway, and use of the network will be a routine aspect of the educational process. On campus this presents no unusual problem and the campus network service can be extended to university residence halls as well. The problem for which we must find a solution is to ensure that we will have in place by then a wide area communications infrastructure supporting affordable high capacity data services that will allow delivery of these essential resources to homes and other off campus locations. I believe the solution can be seen as an evolution from where we are now, through partnerships with new commercial service providers, towards the goal of being able to take full advantage of new developments in communications technology and infrastructure.
Where we are now.
The Berkeley campus community includes around 44,000 faculty, staff, students and researchers. The campus network now includes over 20,000 nodes and is still growing at the rate of around 4,000 per year. Recent plans include support for 100 megabit/second LAN technologies and experimental installation of ATM in support of "video-on-demand" servers.
The Data Communication & Network Services department at Berkeley now supports over 600 modems as part of its campus network services. (See Figure 1) Typically these modems are fully utilized from well before noon until after midnight. If we were to try to scale our modem service to meet projected demand, Berkeley would have many thousands of modems in place by the year 2000. We believe this would be a logistical and management nightmare so what alternative should we foster over the next half decade?
Figure 1: Traditional Method of Connecting to the Campus Network
The use of network communications has become integral to our work, our research, and our teaching and learning activities. The need to maintain access to the network from a variety of off-campus locations has put increasing pressure on the Telecomm and Datacomm service units to provide more and more "dial-in" access points. Most campus members use electronic mail, gopher, and other information resources. Those that use a protocol based connection (PPP) also make heavy use of World Wide Web (WWW) servers and campus-based file servers. Some have file servers at home and would like to use those resources while on campus. A few more technically oriented people have networks of computers in their homes from which they wish to communicate often with the campus network.
Historically most universities first installed dial-in modems as a "free" service, primarily in support of information systems staff and the occasional professor who needed to work at home. Today this service is used by everyone and demand is increasing very rapidly! The primary problem of expanding such a free or "library model" service is that of identifying a funding stream that scales with increasing demand. (Note: Libraries also are looking at this problem and many are considering charging for some services.)
Many campuses are converting to a billed modem service with a modest monthly fee. This incurs additional costs, of course, for account setup, more complex access control systems, billing data collection and processing, and for follow-up on delinquent subscribers.
- Is it reasonable for universities to develop this type of service?
- What alternatives might there be?
Most campus network systems are based on Internet technology (TCP/IP). All large campuses are connected to the world-wide Internet today. In the last year or so, dozens of "dial up" Internet Service Providers (ISP's) have set up business. Many are nation-wide in scope. A few ISP's even will set up modem equipment on your campus and operate it for your community. Most recently Sprint, MCI, and AT&T have joined the growing list of dial-up ISP's.
ISP's have a funding stream designed to allow them to support and expand their capacity (or they won't survive very long!). Therefore, why not "out source" access to the campus network by means of this new class of service providers? (See Figure 2) There are pros and cons to this strategy but we believe that there are very good reasons to adopt this model as quickly as possible.
Figure 2: Commercial Alternative for Accessing the Campus Network
Why would we not use commercial Internet Service Providers?
A major concern on most campuses is total cost to the end-user. If access to the network is integral to the educational program, students will require it. If there is a cost involved, we must be concerned about equity issues and the overall cost of education to the student.
We see a full cost to the university of $60 to $80 per month to support a state-of-the-art modem in a modest sized service pool. Clearly this is too high to pass on to the average campus user directly, especially students. We certainly could adopt a cost model based on statistical "hold times" and prorate a given modem over a number of users. However, as users require more and more time "on line" and the peak use periods overlap, the reality of spreading the cost of a modem across a number of our high demand users is questionable. For the time being however, given a contention ratio of 15:1, a lower bound on monthly costs would be in the $5-10 per month range. Is this significantly less than commercial ISP services?
Commercial ISP accounts are offered in the $10-20 per month range today with a charging model that uses a combination of flat and measured service. A typical example is "$15 per month plus $2 per hour after the first 40 hours." While not nearly as inexpensive as the campus provided service on a "full time month basis" it is quite affordable in absolute dollars to the modest end user. Furthermore, many ISP's allow "free" access during off-peak times giving the determined student much more than the basic service time. It is reasonable to expect that the cost of such services will continue to come down as competition increases. It might be possible to negotiate a "bulk service agreement" with a commercial ISP to bring the cost down for university community members in return for campus guaranteed minimum number of accounts, rebilling of users, or user support functions.
Another concern for most universities is the level of services available from the ISP's. Most campus dial-in services support SLIP and PPP which makes the user's computer appear to be a node on the campus network. Currently many ISP's charge more for this type of service, in part because the support costs are higher and the market requiring this service is still smaller than that for simple "shell" access. University users require SLIP/PPP in order to make full use of information resources such as World Wide Web and other client/server applications. An advantage of negotiating a bulk service contract with an ISP would be the ability to define the service to require SLIP/PPP support.
A related concern is how end users will gain access to campus restricted information. Many current systems or database servers that must restrict access to only "the campus community" use network addresses as the control mechanism. This is done because it is fairly simple to implement but it may not fully achieve either the campus's purpose or the end user's goals. If a user attaches directly to the campus network (i.e. with SLIP/PPP) they will have access to this information. If they attach to an ISP and then attempt to connect to the application, they will be seen as coming from a different network so may be denied access. Clearly the access control method is wrong but there is no good alternative in wide spread use today. This is a problem that must be solved in any case because access control should be based on the user, not on where they are located on the network. A faculty member on sabbatical should be able to gain access to campus resources from a remote location; conversely a visiting scholar may not qualify to access campus licensed resources even though they happen to be "on the campus" temporarily.
Other concerns might include the adequacy of bandwidth between the ISP's modems and the campus network, the increased complexity of problem resolution, and a somewhat greater potential for eavesdropping on traffic. Privacy might also be an issue. As commerce increases on the Internet, ISP's might be tempted to augment revenues by selling information on subscribers preferences to interested marketing services. We must stay closely involved in public policy issues in addition to the purely technical service issues.
There are many good reasons to use Internet Service Providers!
1) Toll free calling
One great advantage that ISP's have that universities would find difficult to implement is "toll free" access for all users. (So-called 800 access is a very expensive option that we find quite unattractive.) In many areas, calls from beyond some distance (approx. 16 miles in the San Francisco Bay Area) are subject to measured service tolls. This creates a serious disadvantage for students, faculty and staff living at a distance from the campus. This disadvantage will be magnified as ISDN becomes deployed for Internet access.
The university could rent space in outlying areas and install phone lines, etc., but is this a reasonable activity for universities to engage in? ISP's typically will have or will create a service point wherever the density of subscribers warrants.
2) Access from anywhere
A secondary advantage of using an ISP is to provide access while traveling. Many faculty and staff take portable computers to conferences and would like to retain access to campus resources. An ISP with nationwide coverage would allow this as easily as if the caller were local. Obviously the access control issues raised above may be a serious problem in this scenario.
3) Competition and variety
A commercial service that faces serious competition will be driven to provide innovative services and motivated to keep costs as low as possible. Newer communications technologies such as ISDN and Frame Relay will be supported as soon as the market allows. Universities may be more limited in their ability to respond to new technologies because of their existing technology base and lack of an adequate funding stream to update it.
An obvious extension of the current dial-in connectivity model (connection of individual workstations) is connection of home networks. Building a LocalTalk or ethernet network is fairly straight forward on a small scale. One computer could serve as a router to connect the entire network to the Internet with a single access line. This is a typical service model for an ISP. Are universities prepared to enter this service market?
4) The "apartment house" model
We see the need to gain access to Internet services penetrating far beyond the university community. We have had inquiries from non-campus-owned living groups that would like to connect a building-wide network to the campus network. While many students live in the building, so do non-students. To connect the entire building seems beyond the scope of serving the university community. However, it fits perfectly the ISP model of connectivity. If we can solve the access control problem posed above, the ISP model could provide very good service to groups of students (and others) at relatively low cost through shared service. Apartment building managers would be motivated to add this service (along with cable TV) in order to attract tenants.
5) Appropriate use of university resources
Clearly an advantage of the ISP model is that most personal use would bypass the university's resource. Furthermore, when students graduate and no longer qualify as part of the "university community" they can retain their Internet identity and access more or less seamlessly. It could be retained for life.
- Where is all this leading?
One way to look at the problem is to ask "what would the ideal world look like in five years?"
The Urban Area Network model.
Just as voice telephone service has become ubiquitous, data network service will become essential to urban life. Learners of all ages will benefit greatly from access to the vast array of new information resources. Commerce will occur via this new pathway. Members of the "university community" will be distinguished not by who provides their "Internet dialtone service" but by their access to university research and learning resources via the network. However clearly we see this vision of the near future, it is just as clear that the university will not literally build the required urban network infrastructure.
As noted above Internet access has already become a commercial service. It is not a service the university should construct throughout the broader community any more than we construct voice dialtone services for the community. Today, the commercial network access services are embryonic and still rely on voice grade modems. We should be working as soon as possible with communications service providers to help define how such services will evolve by the end of this decade into a high capacity, affordable, ubiquitous "urban area network service."
Today the data communications service available to most homes is constrained by the physical characteristics of existing service drops and local loop cabling. We use increasingly complex modems over this media in order to try to get higher speed data transport between home and the network. ISDN is finally becoming available and will support at least 128 Kb/s to the home. However, for the types of applications we envision we will require data rates at the very least an order of magnitude higher. It seems unlikely that this capacity can be provided broadly and economically by current circuit-switched services, regardless of the media.
Existing high capacity services incur relatively high cost for the individual end user. T1, SMDS, or even Frame Relay have high installation and monthly costs and require very expensive termination equipment. Independent of cost, these point-to-point services would be hard to scale to the size required of an urban area. Consider existing voice service equipment sites as "hubs" trying to serve at least 50,000 subscribers at T1 with Frame Relay! We wouldn't build our campus area data networks that way.
Wireless data networks, including packet data services and nationwide data delivery services, are becoming popular. We believe this technology fills a particular need but it seems likely that data rates will remain under 1 Mb/s for wide area services. (Microcellular services have demonstrated data rates in excess of 1 Mb/s but have transmission ranges of less than 20 meters.) Also, since broadcast systems are in effect "shared media," the actual bandwidth available to any particular user may be quite constrained.
A very interesting and relatively new development may offer an opportunity to engineer a cost effective, scaleable, high capacity packet data transport service. Both CATV and voice telecommunications service providers are planning to install new cable plants in major urban areas over the next decade. These likely will consist of "fiber to the neighborhood and coax to the home" (the so-called Hybrid Fiber Coax (HFC) cable plants [1]) with active electronics in neighborhood "hubs." This will support not only voice dialtone and the proverbial "500 channels of mud wrestling" but potentially megabit per second bi-directional transmission speeds for data services. What service model(s) might come as a result of this new infrastructure?
We believe that three key determinants of success will be: subscriber cost, scaleability, and the ability to adapt to a broad potential market. One possible implementation model would be to view the new high speed infrastructure as a packet data transport cloud with information service providers accessible to anyone on the cloud. This "packet data transport service" (PDTS) could be quite similar to Frame Relay or SMDS. With the right engineering this PDTS could be easily scaleable. Scaleability and the fact that the cable plant cost would be recovered largely by the other services should keep the PDTS subscriber cost reasonable.
The PDTS service model would be very similar to that proposed for "video on demand" with information providers attaching to the new urban HFC infrastructure and selling services to any end user similarly connected. One generic type of data service provider would be the Internet Access Provider (IAP). IP packets would be relayed between the user and the IAP via the PDTS. Another major service provider easily could be the university: a provider of learning resources and research information. All that would be required would be a link between the campus network and the urban area network cloud. (See Figure 3)
Fig 3: The University as a Service Provider on the Urban Area Network
The exact details of the technology that might provide high capacity "packet data dialtone" is a subject for another paper. The points that become clearer by envisioning where we might want to be in five years are:
- the Urban Area Network (UAN) won't be built by the university, and
- use of the UAN will cost end users something.
If we accept this conclusion then adopting a strategy to outsource access to the campus network as soon as possible begins to get across the idea that this type of service in their homes will cost end users something - just like cable TV or voice dialtone. It also motivates us to solve the access control problems, and tells us we should be working closely in partnerships with communications technology developers to ensure that emerging technology will serve our needs.
How do we get there from here?
I suggest there are two main areas in which universities might have a significant and effective role. One is in developing robust access control standards and technology so that we can offer proprietary information resources within this new communications environment. The other is in the interface standards and the design and operating characteristics of urban area network technology.
The IETF and other standards bodies as well as commercial developers are working on standard security models for distributed authentication and secure data transport. These must be accompanied by robust authorization mechanisms as well as broadly available implementations for all popular platforms. These technologies will be critical for all network based applications.
The most critical area remains the UAN technology itself. Services being tried today that rely on traditional circuit switched access or shared access to low bandwidth data streams within standa