A Scalable ContentAddressable Network

1
A Scalable Content-Addressable Network

• Sylvia Ratnasamy, Paul Francis Mark Handley,
  Richard Karp, Scott shenker
• UC Berkeley, ATT

SIGCOMM 2001
2
Introduction

• CAN (Content-Addressable Network)
• A distributed infrastructure that provides has
  table-like functionality on Internet-like scales
• System Goals
• Scalable
• Fault-tolerant
• Self-organizing

3
Basic Design

• Basic Idea
• A virtual d-dimensional coordinate space
  (d-torus)
• Each node owns a Zone in the virtual space
• Data is stored as (key, value) pair
• Hash(key) ? a point P in the virtual space
• (key, value) pair is stored on the node within
  whose Zone the point p locates

4
Basic Design

• Each node only need to maintain the information
  of neighbors (for routing purpose)
• Individual nodes maintain 2d neighbors

5
Routing in a CAN

• Greedy algorithm
• If P is within the Zone of current node, return
  (key, value) or failure (if no such key)
• Else forward the query to the neighbor with
  coordinates closest to P
• Average routing path length
• (d/4)(n1/d)
• a d dimensional space
• partitioned into n equal zones

6
CAN Construction

• 1. Bootstrap
• The new node must find a node already in the CAN
• Bootstrap node supplies the IP addresses of nodes
  currently in CAN using DNS
• 2. Finding a Zone
• Randomly choose one point P in the space
• Send a JOIN request destined for P

7
CAN Construction

• 3. Split Zone
• Assign half zone to the new node (X dimension
  first)
• (key, value) pairs from the half zone are
  transferred to the new node
• The new node gets the information of neighbors
  from the previous occupant
• 4. Joining the Routing
• Notices the neighbors the reallocation of space
• Affects only O(d) existing nodes

8
(No Transcript)
9
Node Departure

• Explicit departure
• Hand over its zone to another node to produce a
  valid single zone
• Node failure
• Periodic update messages between neighbors
• Prolonged absence of an update message from a
  neighbor indicates its failure
• A takeover mechanism merges the zone

10
(No Transcript)
11
Design Improvements

• Reduce path length
• Reduce path latency
• Increase fault tolerance
• Increase data availability
• Improved routing performance and system
  robustness vs. per-node state and system
  complexity

12
Design Improvements

• Multi-dimensioned coordinate spaces
• Increase the dimensions of the virtual space
• Reduce the routing path length O(dn1/d)
• Increase the size of routing table O(d)

13
Design Improvements

• Multiple Realities Multiple coordinate spaces
• Improve data availability
• Improve routing fault tolerance
• Reduce the average path length
• Increase the size of the routing table O(r)

14
Design Improvements

• Multiple Dimensions vs. Multiple Realities
• Multiple dimensions shorter path length
• Multiple Realities improve data availability

15
Design Improvements

• Better CAN routing metrics
• When there are more than one choice for
  forwarding, choose the neighbor with the least
  RTT RTT-weighted routing
• Reduce the per hop latency (2440)

16
Design Improvements

• Overload coordinate zones
• Assign more than one node to share the same zone
• Reduce average path length
• Reduce the per-hop latency
• Improve fault tolerance

17
Design Improvements

• Multiple hash function
• Replicate a single (key, value) pair at k
  distinct nodes in the system (use k different
  hash functions)
• Increase data availability
• Reduce the query latency

18
Design Improvements

• Uniform Partitioning
• Using 1-hop volume check
• Achieve load balancing

19
Design Improvements

• Caching and Replication
• Caching Node maintains a cache of the data keys
  it recently accessed
• Replication A popular data is replicated within
  a region surrounding the original storage node
• Increase data availability
• Reduce Query latency
• Achieve load balancing

20
Design Review
d - dimensionality r - number of realities p
number of peer nodes per zone k number of hash
functions size n 218 topology
Transit-Stub latency (100, 10, 1)
21
Conclusions

• CAN provides scalable routing and efficient
  indexing
• CAN is completely self-organizing and
  fault-tolerant
• Open problem DoS attack