Perry E. Metzger wrote:

There are existing deployed solutions like Kerberos that scale far
beyond that and work just fine, and actually address all the things
this protocol seems to leave as an exercise to the reader. And yes,
they're in use in real companies at gigantic scales. (Indeed, Kerberos
is central to Microsoft's technologies these days.)

The example I used was only illustrative.  SKSML allows for 2^64 keys
per server, 2^64 servers per domain and 2^64 unique domains on the
internet.  The GlobalKeyID (GKID) of a key within a Symmetric Key
Management System (SKMS) is defined to be a triple consisting of the
concatenation of the unique domain ID, the server ID and the key ID (DID-SID-KID). So GKID's can range from "1-1-1" all the way upto
"(2^64-1)-(2^64-1)-(2^64-1)"; so it is fairly scalable.

That said, Kerberos clearly has the benefit of 20+ years of research
and use in the field.  However, there are two fundamental differences
between SKSML and Kerberos, IMHO:

1) The design goals for Kerberos were very different from SKSML.  The
   former solves the problem of network-authentication in the face of
   insecure hosts/networks, while the latter focuses on long-term key
   management.  That they both use very similiar concepts & components
   to deliver quite different benefits to applications is testament to
   the strength and flexibility of the underlying components rather
   than to a weakness of SKSML.

2) AFAIK, Kerberos clients cannot deliver their stated benefit (network
   authentication) in the absence of the network or the Kerberos server.
   SKSML is designed to allow disconnected EKMI clients to continue
   providing cryptographic services to applications even in the absence
   of the network or the key-management server.  However, it does
   require that the Symmetric Key Client Library (SKCL) have connected
   to the Symmetric Key Services (SKS) server at least once before it
   can use this capability.

Arshad Noor
StrongAuth, Inc.


---------------------------------------------------------------------
The Cryptography Mailing List
Unsubscribe by sending "unsubscribe cryptography" to [EMAIL PROTECTED]

Reply via email to