Negotiating Security Services

With the uncontrolled explosion of incompatible security mechanisms available today comes a significant, unaddressed, problem of how two machines can choose a suite of mechansisms for a particular type of communication.

The degrees of freedom in the problem include the algorithms used, modes of the algorithms, where the services

Why You Must Consider Protection Suites, Not Individual Mechanisms

There are several reasons why you should - insecure - harder to manage - more overhead

The Need for System-Wide SPIs

There is currently a push in the IESG to make a SPI local to a particular transform. This is a huge mistake. As noted above, the transforms that are being applied all over the place are NOT independent. Since we must treat a stack of services as a whole, but have to treat each SPI as transform-specific value, we are forced to introduce a notion of "meta-SPI" for no good reason.

Allowing a security manager (e.g., a negotiation server) to manage the SPI space for all transforms is a much cleaner solution. There will be less duplicated functionality in each transform, and the security manager will be simpler as well.

Links to ISAKMP and Related Resources

ISAKMP Documentation

The latest Oakley Draft (Draft 1)

The latest ISAKMP/Oakley Resolution Draft (Draft 1)

ISAKMP Implementations

SKIP - The S

Group Key Management

GKMP Architecture

GKMP Specification

Skeme

Secure DNS The most likely short-term route for certificate management.

Quality of Service Research

I believe that security service negotiation should be simply an aspect of overall QoS negotiation, and I'm working toward that end. Here are some pointers to current QoS work.


Jeff Turner
sjt@cs.utah.edu