SlideShare a Scribd company logo
Scaling systems securely: challenges and
risks
William Whyte, Chief Scientist
2016-03-16
About Security Innovation
• Authority in Software Security
– 15+ years research on vulnerabilities
– Security Testing methodology adopted by Adobe,
Microsoft, Symantec, McAfee, and others
– Security partner for Dell, Microsoft, Cisco, HP, IBM, PCI
SSC, FS-ISAC, NXP, and others
– Authors of 16 books, 4 co-authored with
Microsoft
– 9 Patents
About me
• Standards
– Chair of IEEE P1363, vice-chair of IEEE P1609
• Large-scale design
– Member of design team for Secure Credential Management System for V2V deployment
• Deployment
– Working with New York City Pilot Deployment to develop Security Management
Operating Concept
• Cryptography
– Member of research team developing NTRU and related cryptographic technologies
• Major professional focus: How can we get a 300 million device system to work
securely?
Outline
• Growing systems securely
Outline
• Growing systems securely
• Securely
– Setting them up
– Maintaining them
– Data management and privacy
Outline
• Growing systems securely
• Securely
– Setting them up
– Maintaining them
– Data management and privacy
• Growing
– Across number of devices
– Around the world
– Through time
Our main setting: Vehicle-to-Anything (V2X)
Illustrations from https://www.itsconnect-pc.org/en/about_its_connect/service.html
Our main setting: Vehicle-to-Anything (V2X)
• Peer-to-peer applications
– Multiple different suppliers
• “Centralized” security management
– Not always contactable
– May be different across applications
– May be different across countries
• Long-lived devices
– Cost sensitive
– Limited central connectivity
– Limited P2P bandwidth
• Huge scale
– 300 million cars on the road in the US, mandate coming
The biggest constraint on scalability
• People
The biggest constraint on scalability
• People who have to get it right
People taking responsibility is the bottleneck
• Individual decisions made centrally by humans don’t scale
– Lumpy
– Hard to plan for
– Slow
• Edge decisions made by humans can scale so long as there aren’t too many of
them
• Scalable decisions made centrally by humans do scale
– Write this code, push this update, …
• Systems scale successfully when decisions made by humans can be scaled
• Scaling is a risk when those decisions are wrong
• How can we leverage central decisions to make sure my car gets good data from
your car?
Peer-to-peer
• How can my car be sure that your car’s data is good?
• Bad faith / poor execution
• Need 300m devices that are:
– Authenticated
– Certified
– Well-implemented
– And have 1000+ certificates per year
• … without needing 300 billion human decisions
V2V Security Credential Management System
• Standard PKI plus…
• Large-scale revocation
• Privacy protection
• Lots of devices
• How do we manage central
decisions made by people?
– Enrolment v authorization certs
– Device certification
– Revocation and misbehavior
investigation
Enrolment v authorization
• Enrolment certificate at the start of
device’s lifetime
– Used only for SCMS communications
• Authorization certificate refreshes
during device’s lifetime
– Used for peer communications
• Down to 17m decisions per year!
• Now the problem is to make sure
the Enrolment CA makes the right
decisions
Device certification
• Device is correctly implemented
Device certification
• Device is correctly implemented
– We have the requirements right
Device certification
• Device is correctly implemented
– We have the requirements right
– It is properly tested
Device certification
• Device is correctly implemented
– We have the requirements right
– It is properly tested
– It will stay secure
Device certification
• Device is correctly implemented
– We have the requirements right
– It is properly tested
– It will stay secure
• If we can guarantee this then we only need
type certification
– But…
Device certification issues
• Testing against requirements is expensive
– $100,000 for Common Criteria conformance
– Bug fixes could require recertification
• Requirements change
– Attacks evolve
• Different labs have different standards
– … because different countries have different standards
• US/EU/Australia Harmonization Task Group 6 on Intelligent Transportation Systems Security
– US status of type certification for V2V devices is still open
• Ability to scale is a problem of governance, not a problem of technology
Disaster recovery in large systems
• Removing bad actors
– Privacy v security
• Repairing faults
Removing bad actors
• Cars can report suspicious
messages
– “Ghost vehicles”
• Advantage: misbehavior can be
detected
• Disadvantage: innocent users can
be tracked
• How to enable misbehavior
detection without widespread
tracking?
– Careful data management processes
• Is this a universal problem?
Removing bad actors: and finally…
• Can a removal decision be
legally challenged?
• If so, this becomes one of
those human decisions we
were trying to
minimize
Outline
• Growing systems securely
• Securely
– Setting them up
– Maintaining them
– Data management and privacy
• Growing
– Across number of devices
– Around the world
– Through time
Future threats & mitigations
• Quantum computers
– Can break all current popular public-key
cryptography
• Threat to application messages
– Research going on through, e.g.,
http://pqcrypto.eu.org/
• Threat to secure update
– Secure post-quantum signatures exist but…
– Currently no hardware support
Conclusions
• Minimizing human decisions
– Device certification
• International agreement on requirements
• International agreement on conformance testing
culture
– Clear decision criteria for revocation
• Proper data management
– Preserve privacy for
• General users of the system
• Innocent users accidentally included in misbehavior
reports
• Users suspected of violations?
• Secure update
– Crypto algorithm agility
– As little reliance as possible on hardware for
application messages
– Hardware support for post-quantum secure
update
When building a secure, large-scale, peer-to-peer, privacy preserving,
international, long-lived, poorly connected system, pay attention to…

More Related Content

Scaling Systems Securely: Challenges and Risks

  • 1. Scaling systems securely: challenges and risks William Whyte, Chief Scientist 2016-03-16
  • 2. About Security Innovation • Authority in Software Security – 15+ years research on vulnerabilities – Security Testing methodology adopted by Adobe, Microsoft, Symantec, McAfee, and others – Security partner for Dell, Microsoft, Cisco, HP, IBM, PCI SSC, FS-ISAC, NXP, and others – Authors of 16 books, 4 co-authored with Microsoft – 9 Patents
  • 3. About me • Standards – Chair of IEEE P1363, vice-chair of IEEE P1609 • Large-scale design – Member of design team for Secure Credential Management System for V2V deployment • Deployment – Working with New York City Pilot Deployment to develop Security Management Operating Concept • Cryptography – Member of research team developing NTRU and related cryptographic technologies • Major professional focus: How can we get a 300 million device system to work securely?
  • 5. Outline • Growing systems securely • Securely – Setting them up – Maintaining them – Data management and privacy
  • 6. Outline • Growing systems securely • Securely – Setting them up – Maintaining them – Data management and privacy • Growing – Across number of devices – Around the world – Through time
  • 7. Our main setting: Vehicle-to-Anything (V2X) Illustrations from https://www.itsconnect-pc.org/en/about_its_connect/service.html
  • 8. Our main setting: Vehicle-to-Anything (V2X) • Peer-to-peer applications – Multiple different suppliers • “Centralized” security management – Not always contactable – May be different across applications – May be different across countries • Long-lived devices – Cost sensitive – Limited central connectivity – Limited P2P bandwidth • Huge scale – 300 million cars on the road in the US, mandate coming
  • 9. The biggest constraint on scalability • People
  • 10. The biggest constraint on scalability • People who have to get it right
  • 11. People taking responsibility is the bottleneck • Individual decisions made centrally by humans don’t scale – Lumpy – Hard to plan for – Slow • Edge decisions made by humans can scale so long as there aren’t too many of them • Scalable decisions made centrally by humans do scale – Write this code, push this update, … • Systems scale successfully when decisions made by humans can be scaled • Scaling is a risk when those decisions are wrong • How can we leverage central decisions to make sure my car gets good data from your car?
  • 12. Peer-to-peer • How can my car be sure that your car’s data is good? • Bad faith / poor execution • Need 300m devices that are: – Authenticated – Certified – Well-implemented – And have 1000+ certificates per year • … without needing 300 billion human decisions
  • 13. V2V Security Credential Management System • Standard PKI plus… • Large-scale revocation • Privacy protection • Lots of devices • How do we manage central decisions made by people? – Enrolment v authorization certs – Device certification – Revocation and misbehavior investigation
  • 14. Enrolment v authorization • Enrolment certificate at the start of device’s lifetime – Used only for SCMS communications • Authorization certificate refreshes during device’s lifetime – Used for peer communications • Down to 17m decisions per year! • Now the problem is to make sure the Enrolment CA makes the right decisions
  • 15. Device certification • Device is correctly implemented
  • 16. Device certification • Device is correctly implemented – We have the requirements right
  • 17. Device certification • Device is correctly implemented – We have the requirements right – It is properly tested
  • 18. Device certification • Device is correctly implemented – We have the requirements right – It is properly tested – It will stay secure
  • 19. Device certification • Device is correctly implemented – We have the requirements right – It is properly tested – It will stay secure • If we can guarantee this then we only need type certification – But…
  • 20. Device certification issues • Testing against requirements is expensive – $100,000 for Common Criteria conformance – Bug fixes could require recertification • Requirements change – Attacks evolve • Different labs have different standards – … because different countries have different standards • US/EU/Australia Harmonization Task Group 6 on Intelligent Transportation Systems Security – US status of type certification for V2V devices is still open • Ability to scale is a problem of governance, not a problem of technology
  • 21. Disaster recovery in large systems • Removing bad actors – Privacy v security • Repairing faults
  • 22. Removing bad actors • Cars can report suspicious messages – “Ghost vehicles” • Advantage: misbehavior can be detected • Disadvantage: innocent users can be tracked • How to enable misbehavior detection without widespread tracking? – Careful data management processes • Is this a universal problem?
  • 23. Removing bad actors: and finally… • Can a removal decision be legally challenged? • If so, this becomes one of those human decisions we were trying to minimize
  • 24. Outline • Growing systems securely • Securely – Setting them up – Maintaining them – Data management and privacy • Growing – Across number of devices – Around the world – Through time
  • 25. Future threats & mitigations • Quantum computers – Can break all current popular public-key cryptography • Threat to application messages – Research going on through, e.g., http://pqcrypto.eu.org/ • Threat to secure update – Secure post-quantum signatures exist but… – Currently no hardware support
  • 26. Conclusions • Minimizing human decisions – Device certification • International agreement on requirements • International agreement on conformance testing culture – Clear decision criteria for revocation • Proper data management – Preserve privacy for • General users of the system • Innocent users accidentally included in misbehavior reports • Users suspected of violations? • Secure update – Crypto algorithm agility – As little reliance as possible on hardware for application messages – Hardware support for post-quantum secure update When building a secure, large-scale, peer-to-peer, privacy preserving, international, long-lived, poorly connected system, pay attention to…