Low Latency Information Slicing

1
Low Latency Information Slicing

• Presentation by Maya Zuhl, Shravya Konda, Liping
  Liu

2
Content

• Project motivation
• Background
• Algorithm in details
• Evaluation (latency, security)?
• Conclusion
• Future work

3
Project Motivation

• P2P overlay networks
• Security
• Quick delivery
• New interesting and very useful field of computer
  networks

4
BackgroundInformation Slicing

• Two stages
• Routes setup
• Data transfer
• Confidential messages delivery over disjoint
  paths to the destination
• Each message is scrambled
• Split into d pieces
• Sent over d disjoint routes

5
BackgroundInformation Slicing
Problems
Aim

• Provides confidentiality without asymmetric
  encryption
• Does not need a trusted third party
• Does not require public key infrastructure
• Does not impose heavy overhead on the nodes
• Provides reasonable anonymity and security

The routes are chosen arbitrary Requires access
to at least two internet connection points that
have secure connection between them Makes a lot
of assumptions that are difficult to meet in
reality
6
BackgroundNetwork Coordinates - Vivaldi

• Decentralized
• Adaptive (inspired by a system of springs)?
• Low overhead for the network
• Allows to estimate RTT in a very precise way
• Uses triangular inequality violation to predict
  the actual RTTs between the hosts
• Adjusts to various system coordinates like
  Euclidean 2 or 3 dimensional system

7
BackgroundNetwork Coordinates - PeerWise

• Finds neighbour nodes with the lowest latency
  possible
• Establishes mutually beneficial peering
  relationship
• Proves that usually one-hop detour is enough to
  sufficiently lower the total path latency

8
BackgroundSimilar Approaches

• One technique for securely delivering data in
  structured overlays is to increase the number of
  disjoint paths among peers A Novel Methodology
  for Constructing Secure Multipath Overlays by
  Marc Sánchez Artigas et al.
• Redundant routing technique as a mean to defend
  against message dropping by Miguel Castro et
  al.
• Disjoint (independent) lookup paths for DHT based
  overlay networks by Mudhakar Srivatsa and Ling
  Liu

9
LLIS

• Low Latency Information Slicing
• - fast and basic security protocol
• Assumptions
• - no global attacker who can snoop on all
  links
• - limited malicious attackers
• - every source has a pseudo source.

10
LLIS Design

• Two procedures
• Setup the routing graph
• - source routing is done
• - distribute the symmetric keys
• Actual data transmission by the sources
• - how is the packet sliced

11
Routing Setup

• Find overlay paths between the sources and
  destination.
• Slicing only at the source
• d number of slices
• L path length
• Transfer half of the slices to the pseudo source
• Choosing overlay paths
• - Start at the source
• - Run Peerwise at each node on the path
• - choose d next best hops for each node
  based on latency metric.

12
Packets Propagation Example
4
3
8
5
1
1
1
1
2
1
2
6
8
1
3
1
1
7
4
1
1
1
1
3
5
4
2
2
1
1
13
Information Slicing and Reassembly

• Every packet is sliced into d parts
• Each part is multiplied by a row of a random
  matrix A
• Every such slice is sent along a node disjoint
  path
• Same Flow-id to all slices of the same packet
• Reassembling
• - flow-id

14
Resilience to Node Failures

• Basic Idea
• - original vector is of dimension d
• - multiply by random invertible matrix A (d x
  d)?
• - modified vector is of dimension d
• Redundant Routing
• - choose top d lowest latency paths
• No data loss
• - d independent slices

15
Evaluation

• Latency
• - data transmission from sending time to
  receiving time
• - routing setup
• Security
• - against packet dropping
• - traffic analysis attack
• - packet changed by attacker

16
Simulation Environment

• Normal encryption approach Traffic configuration
• - CBR traffic
• - Packet size1000
• Normal encryption approach
• - encrypt the message with key
• - randomly select next hop
• LLIS simulation configuration
• - d2
• - d3, provide redundancy
• - L4, upper bound
• - N total no. of overlay nodes, more than
  20 nodes
• - select d lowest next hop

17
Comparison with Normal Encryption Approach

• Green line latency of LLSI
• Colourful line normal encryption approach

18
Latency Analysis

• Routing setup latency
• - required only once
• - assume coordinate information available for
  every node to make simulation simpler
• Results on the way

19
Latency Analysis

• Improvement near 48
• - thats what fast means
• Improvement from
• - node disjoint path
• - d lowest latency path (not the best perfect
  path)?
• - no encryption in the middle
• - cost of encoding and decoding time is low
• (d5, 60us)?

20
Security Evaluation

• Malicious attackers collusion
• (relay choosing different AS and lowest delay)?
• Denial-of-service attack
• (possibility of increasing the size of network)?
• Node churn (redundant routing)?
• Traffic analysis attack
• - End-to-end time analysis (relay choosing)?
• - Packet size analysis (constant size)?

21
Conclusion

• Trade off between latency and security
• Low latency and reasonably secure approach
• Does not require public keys infrastructure
• Resilience to node churn
• Copes well with basic security attacks

22
Future work

• Many strict assumptions
• Routing setup analysis
• Parameters effect on performance

23
Questions?