NetworkSecurity Talking Points

1
Network/Security Talking Points

• ECI Workshop
• NSF
• 6-7 Dec 2004

2
Major Topics for Discussion

• Networking Trends Bigger, Faster Cheaper but
  its the software, stupid
• Security Concerns
• User Identification
• Role-based Authorization
• Data integrity
• Data Security
• Privacy

3
Networking in the 21st Century

• National 10-gigabit research networks
• TeraGrid, National LambaRail
• Internet2 backbone to go beyond 10 Gigabits
• International connections at 10 Gbits and
  growing
• Shared/distributed datasets can be quite large
• Networking and application software have a long
  ways to go to effectively utilize this this
  resource

4
National Lambda Rail

• Consortium of GigaPOPs that collectively own
  1000s of miles of fiber
• Multiple 10-gigabit networks running on this
  fiber (DWDM)

NLR Map Source John Silvester, Dave Reese, Tom
West, CENIC
5
Driving Observations

• Aggregate carrying capacity of fiber is doubling
  faster than yearly
• DWDM (long-haul), CWDM (Metro, Campus)
• Each fiber carries multiple signals
  differentiated by color
• System network interface increases by O(10)
  every 5 years
• This is on Moores curve, not on the fiber curve
• Over the next decade, the external bandwidth to
  a collection of machines (cluster) roughly
  matches their aggregate BW
• Value of the external network changes
• Aside NIC bandwidth approaches memory bandwidth

6
Reality The Clogged (and ossified) Internet
File Transport, NASA EOSDIS
Source Bernard Minster, SIO, UCSD
7
Critical Networking Challenge

• Observe that networks are getting significantly
  faster
• Learn to design software for this future
  environment
• MIT Athena Project took this exact approach with
  X-Windows

8
Security

• User Identification
• Globus team proposed 10 years ago that public key
  cryptography and user credential management was
  an essential building block for mutually
  authenticating single sign on grids (GSI)
• Right technology
• Too hard for users with the current state of
  tools (this is improving)

9
How Single Sign On Works (Abbreviated)

• User requests a public/private key pair from a
  certificate authority (CA)
• CA issues pair to user, records the footprint and
  makes the user responsible for management
• User creates a grid proxy (time limited) from
  private key. Proxy can be validated with the
  users public key.
• Proxy is transferred to a site as the identity of
  the user
• If the proxy is valid
• If the site trusts the issuer of the users
  certificate
• If the site can match the valid identity to a
  local account
• If the local account is in good standing
• Then, the user is signed onto the grid resource

10
Identity Management is Step 0

• Real-world problems
• Explicit certificate management by users is
  untenable
• Users lose passwords
• Users lose private/public keypairs
• Users mistakenly transmit passwords in the clear
  because private key is on a shared resource (eg.
  NFS share).
• Sites read too much into what a certificate
  Certifies
• Emerging common solution
• A grid certificate bank holds private/public
  keypairs
• Using only a small number of access mechanisms,
  the bank will generate a proxy on behalf of the
  user (e.g. MyProxy or CAS)
• Users only see username/passwords
• This is only the initialization step, Grids still
  have to understand what roles a particular user
  has.

11
Identity Management Challenge 1

• It is easy to build Certificate Authorities (eg.
  One for NEON, one GEON, one for Teragrid, )
• It is more difficult to get other sites to accept
  the a foreign CA signing policy
• Identity Trust/Transformation Systems (Eg.
  Shibboleth) can ease this.
• ?? For all grid based science
• Build or Buy a CA?
• Second challenge, what happens when a user has
  multiple certificates? (E.g. which passport does
  a dual citizen use to enter a country)
• Third Challenge what do you read into the
  identity provided by a certificate?

12
Authorization

• Identity just says who, not what is allowed
• Role-based authorization is one essential
• A dearth of tools of exist in this area

13
Data Integrity

• How do you validate data that resides in an
  archive
• Do not believe that magnetic storage systems (eg.
  Disk) dont mangle bits . bit rot is real.
• How do you validate data that is coming from
  sensors ?
• How do you provide data provenance for derived
  data?

14
Data Security

• End-to-End Encryption is the only type of
  encryption that can be reasoned about
  (transmission security)
• How do you audit who has accessed/changed data?
• User (and machine) authorization (eg. Derived
  from GSI credentials) is critical
• Can you watermark digital data so that the
  original source is embedded in the complete set

15
Data Privacy

• Can outsiders determine who has accessed what on
  the grid?