Much research and development effort has been spent on creating new Internet applications, protocols and network-layer services, with limited success. However, from a user perspective, the major problems in the Internet today are not the lack of applications or the need for sensor and ad-hoc networks. Rather, as network technology matures, the challenges are trustability, maintainability and reliability. The solution cannot involve turning all Internet users into trained network and system administrators or private detectives. In the talk, I will more questions than provide answers.
Drawing on examples from IETF protocol design efforts, I will try to illustrate this changing landscape of protocol design and its challenges for network-related research. The network environment has changed fundamentally in the last few years, where scaling has principally enabled a tiny number of malicious and criminal elements to impose huge costs on the vast majority of users.
The current mechanism, characterized by security patches, the proliferation of access secrets and unexplainable denials of service to legitimate users and usages, cannot scale. We need to start considering alternatives. Such alternatives might involve the transition from a global village back to more controllable environments and a fundamental switch from a model which allows compromised or malicious hosts to point ever heaver artillery at random victims to "permissions-based" networking that minimizes the damage that untrusted hosts can inflict.
Another interesting question is where the narrow waists of the protocol stack hourglass will emerge. For example, will we see the convergence into a very small number of general-purpose protocols such as HTTP (with SOAP), SIP and SMTP or the proliferation of application-specific protocols?
Maintenance cost for user applications is currently high, with excessive manual configuration. Little attention has been paid to creating light-weight protocols, as measured not in CPU cycles, but in user configuration expense.
Finally, reliability requires thinking beyond traditional network management approaches. We have very little insight into why network connections fail, who is to blame and where we can make the most difference.
|