7.9 Where the Field Is Headed

Much work is being done to enhance the security of networks. Research by vendor companies will lead to more flexible and secure boxes, while more fundamental research will look into the fundamental problems of networking: authentication, access types, and authorizations. A particular problem of security in networks is one of speed: As the speed, capacity, bandwidth, and throughput of networks and network devices continue to increase, security devices have to keep pace, which is always a challenge.

A second security challenge with networks is ubiquity: As automobiles, cell phones, personal digital assistants, and even refrigerators become network enabled, they need security. The need for a firewall for a cell phone will become apparent the first time a cell phone is subject to a denial-of-service attack. Once again, security will be called upon to protect after a product is in use.

Joshi et al. [JOS01] present seven different models that could be used for access control in networked applications. These models include the decades-old mandatory and discretionary access control, about which literally hundreds of research results have been published, and more recent task- and agent-based models. The article is an excellent analysis of the models and their applicability in different network situations. But the article clearly shows the immaturity of network security if after three decades into networking we still need to analyze which security models are appropriate for networking.

Protocol development continues as new networked applications arise. The challenge is to ensure that protocols are scrutinized for security flaws and that security measures are incorporated as needed. An example of a new protocol that addresses interesting security needs is Stajano and Anderson's "resurrecting duckling" protocol [STA02]. This protocol invents the concept of a "secure transient association," to describe a connection that must be created, acquire security properties, operate , and terminate, perhaps passing those properties to other entities. This work is a good example of development of the protection model before the need arises.

The firewall technology has matured nicely in the past decade . The pace of innovation in firewalls has slowed, and it seems as if freestanding firewalls have gone about as far as they can. But we can expect to see more firewall features incorporated into applications, appliances, and devices. The personal firewall to protect a single workstation is a good example of how security technology is extended to new domains of use.

Intrusion detection systems have a much longer history than firewalls, but they also have further to go. Interesting new work is underway to define "good" or "safe" behavior and to restrict access rights. (See, for example, [KO97, FOR96].) The next big challenge for IDS products is to integrate data from more sources to be able to infer a threat picture from many individually insignificant clues.

Denning [DEN99] has done very thorough and thoughtful study of the potential for misuse of the Internet for political purposes or terrorism. Because it is becoming so much a part of the essential infrastructure for commerce, education, and personal interaction, protection of the Internet is very important to society. But because of its necessarily open structure, protecting it is often inconsistent with promoting its use. Denning carefully explores these issues, as well as the possibility of using the Internet for harm.

The security of mobile code will become a larger issue as remote updates and patches and downloading of agents continue to increase. The classic security problem of demonstrating assurance is exacerbated by the anonymity of networking. Rubin and Geer [RUB98] provide a good overview of the field. Proof-carrying code [NEC96], code signing, type checking [VOL96], and constrained execution [GON96, GON97] are possibilities that have been proposed. (Interestingly, these approaches build on the use of the compiler to enforce security, a technique first suggested by Denning and Denning [DEN77].)