Protecting Voice Telephony in the 3GPP IMS

1
Protecting Voice Telephony in the 3GPP IMS

• Presented by Gabriel Andrews
• Supervised by Neco Ventura
• University of Cape Town

2
Networks of the Future

• 3GPP is standardizing an all-IP-Network
• Differing devices with differing resources all
  communicating with each other
• VoIP will be a cheap, common-place service that
  will utilise this infrastructure
• Calls between soft-phones, PSTN stations, mobile
  handsets and computers

3
Tragedy of the commons

• Economics shows that when actors with different
  capabilities use the same resource, abuse occurs
• How do we protect such systems economics
  advises us to apply taxes
• Tax used to ensure that total utility of the
  network is maximised

4
Tragedy of the digital commons

• In a communication system a users attention
  endowment is the precious resource
• Actors compete to for users attention try to
  get user to respond to their communication
  attempts
• Under certain circumstances abuse occurs

5
Tragedy of the digital commons

• This abuse is commonly known as SPAM in email, in
  VoIP networks it is called SPIT
• Various solutions have been proposed within the
  context of email SPAM
• These are the basis for SPIT prevention strategies

6
Problem context

• Wireless VoIP network
• Various devices with varying capabilities i.t.o.
  CPU speed, memory speed and space
• Loosely defined identities
• Need to emulate existing mode of voice service
• No a-priori information about validity of message

7
Assumptions

• Spammers are sophisticated users who have access
  to powerful machines and large bandwidths
• The spammers goal is to create as many
  unsolicited sessions as possible, as quickly as
  possible
• Spammers initiate a large amount of one-time
  sessions

8
Assumptions

• Legitimate users access service via mobile
  handset
• Users initiate repeated sessions
• Duration of session is greater than the session
  setup time

9
Proof of effort

• Solutions known as proof of effort
• Users pay some tax to use service
• Tax can be time or money
• How does this apply to this context

10
Proposed Solution

• Memory access speeds vary less significantly
  across devices than processor speeds
• Develop algorithm whose processing time is memory
  bound
• First proposed by Adabi et al. and then by Dwork
  et al.

11
Memory Bound Algorithms

• Based on a directed acyclic graph (DAG)
• We control starting point, width and height of
  tree
• Must be too large to store in cache
• Program execution relies on random access to
  matrix in random steps

12
Memory Bound Algorithms

• Results in cache misses and hence cache swap-outs
• Users return top hash or a result when some
  condition is satisfied
• Central authority then verifies the response

13
Session Setup Example
14
Result
15
Conclusion

• Memory bound computational puzzles used as tax
• Limits the amount of calls that can be made
• We are punishing a certain behaviour and thus
  have minimal impact on legitimate users
• Hard to compute, easy to verify

16
Questions

• ???

17
(No Transcript)
18
Operation of algorithm
T
TB