A Methodology for Analyzing the performance of Authentication Protocols

1
A Methodology for Analyzing the performance of
Authentication Protocols
Alan Harbitter Daniel A. Menasce
Presented by Rob Elkind
2
Outline

• Introduction
• Kerberos and extensions
• Kerberos with Proxy
• Methodology
• Simulations Multiple Realm and Mobile with
  proxy
• Conclusion

3
Introduction

• Use of new modeling methodology for analyzing
  authentication protocols
• Closed queuing network model
• Two Kerberos examples will be tested
• Designed to explicitly model performance new
  protocol design including asymmetric and
  symmetric encryption

4
Kerberos Overview
5
Kerberos Realms

• Kerberos realms - networked collection of
  workstations, servers, and a single master KDC
  which must
• 1. maintain a database of matching user IDs and
  hashed passwords for registered Kerberos users
• 2. maintain shared secret keys with each
  registered application server
• 3. maintain shared secret keys with remote KDCs
  in other realms
• 4. propagate new or changed secret keys and
  database updates to slave KDCs.

6
Public Key Cryptography

• Increase scalability
• Smaller key shared space n2 vs. n for n users
• Improved Security
• Proposals
• PKINIT (core specification)
• PKCROSS
• PKTAPP

7
PKINIT Overview
8
PKCROSS Overview
9
PKDA Overview (PKTAPP)
10
Proxy server with Kerberos

• Isolate client and server for security purposes
• Offload processing from mobile host or network
• IAKERB
• Charon

11
Methodology

• Build model
• Validate
• Change parameters
• Analyze results
• Add What ifs

12
Modeling Topology multiple-realm
13
Validation of Model
14
What-If Analyses

• Vary input parameters to reflect various real
  world conditions
• Reflects sensitivity to various operational
  environments
• Gives insight into general performance
  characteristics of the protocol design

15
Analysis of Public-Key-Enabled Kerberos in Large
Networks

• Compare PKTAPP and PKCROSS
• Simulate using closed queuing network model
• Use skeleton software to model real world
  protocol
• When is it more efficient to authenticate to a
  central KDC than to individual application
  servers?

16
(No Transcript)
17
(No Transcript)
18
PKCROSS vs. PKTAPP
19
What-Ifs Results
20
Analysis Of Public-key-enabled Kerberos InMobile
Computing Environments

• Reduce the number of public/private key
  operations performed on the mobile platform.
• When a proxy is used, maintain the option to
  preserve the encrypted data stream through the
  proxy.
• Retain the standard Kerberos formats for messages
  sent to the KDC and application server.
• Preserve the semantics of Kerberos.

21
M-PKINIT
22
MP-PKINIT
23
Modeling Topology MMP-PKINIT

• Can use same model as before
• Substitute a mobile client for client
• Wireless network for LAN
• Proxy server for local KDC
• Adjust branching probabilities to reflect new
  model paths

24
Model Results
25
Model vs. Simulation
26
What-If Analysis
27
More What-Ifs
28
Conclusions

• Closed queuing model with class switching is a
  useful tool for analyzing performance in security
  protocols supports wide range of operating
  conditions
• Skeleton implementation is a good way to work
  with new ideas that may not be operational yet
• PKCROSS outperforms PKTAPP for authenticating to
  more than one server
• Proxy server benefits 2G speeds but not 3G speeds

29
Thoughts

• Well written and presented, clear and detailed
• Good procedural methodology
• Would be nice to see What-Ifs done on the test
  bed and compared to model as well
• Skeleton makes assumptions that may alter results
  when performed with real implementation

30
Questions?