Discovery and Consistent Hashing

1
Discovery and Consistent Hashing

• Chris Cabanne ccabanne_at_uci.edu

2
Discovery and Cyber Entity Lookup

• Primary challenge efficiently locating services
  (i.e. Cyber-Entities) across BioNet platforms in
  a decentralized manner.
• Methods
• Relationship Discovery
• Indexing

3
DNS and the Internet

• DNS domain name space.
• Host Name to IP address Mapping.
• i.e. www.uci.edu ? 128.200.222.100
• Indexing is done in a hierarchical manner.
• Request --gt Root Server ? Master Name Server
  (.edu) ? return address.
• Addressed used in network layer routing.
• Constraints
• Semi-static network IP addresses do not move
  geographically
• Possible point of failure Root DNS servers.

4
DNS and BioNet

• Platform (semi-static) may leave or rejoin
  network but stays in a static geographic
  location.
• CE (dynamic) may migrate from one platform to
  the next.
• How do we keep track of CEs location?
• CE may migrate.
• CE may die.
• There may be several cached copies of a Cyber
  Entity.
• Locating a unique CE vs. locating a CE that is a
  member of a group.

5
Consistent Hashing

• One possible solution Consistent Hashing
• Originally a distributed caching technique to
  relieve Hot Spots on the Internet.
• Benefits
• Flexible Naming No imposed naming structure of
  CE
• Decentralization No central control.
• Availability Allows for the location of
  migrating CE

6
Some networks protocals that implement consistent
hashing

• OceanStore http//oceanstore.cs.berkeley.edu/pub
  lications/papers/abstracts/silverback_sosp_tr.html
  
• Chord http//www.pdos.lcs.mit.edu/chord/
• Grid http//www.pdos.lcs.mit.edu/grid/

7
Consistent Hashing Claims

• Claim In an network with N nodes (N BioNet
  platforms) and K keys (Cyber Entities)
• Every Node responsible for holding O( K / N )
  keys.
• Resolves all CE lookups via O( log N ) messages.
• Each Node only maintains routing information to
  O( log N ) other nodes.

8
Consistent Hashing Claims (Continued)

• Claims (Continued)
• Nodes joining or leaving network maintain
  routing information between nodes in O( log² N )
  messages.
• Guarantee correct lookup of CE with only one
  piece of correct routing information.
• Ok, but how does it work?

9
Consistent Hashing

• Basic Idea given a key, map the key onto a node.
• All nodes and keys are assigned an m-bit
  identifier using a base hash function (like
  SHA-1)
• A nodes identifier (must be unique) is chosen by
  hashing the nodes IP address.
• A keys identifier is produced by hashing the key.

10
Assigning Keys to Nodes

• Identifiers are ordered in a circular number line
  0 to 2m - 1.
• Key k is assigned to the first node whose
  identifier is equal to or follows ks identifier
  in the number line.

11
Example of number line
m 3 There are three nodes 0, 1, and 3. There
are three keys 1, 2, and 6 Key 1 is located at
node 1, key 2 is located at node 3, key 6 at node
0.
6
1
0
7
1
2
6
3
5
4
2
12
Scalable Key Location

• Only a small amount of routing information needs
  to be contained at each node for correctness.
• If each Node keeps a pointer to the next node on
  the number line, then the network is fully
  connected every node can reach every other node
  (although slowly).
• To speed things up, every node contains routing
  data to a small amount of other nodes.

13
Node Routing table

• Each node n contains a table that has at most m
  entries (the bit length of identifier hash). This
  is because there can be at most 2m unique nodes.
  
• The ith entry contains the identity of the first
  node, s, that succeeds n by at least 2i-1.
  Calculated as ( n 2i-1 ) 2m, 1 lt i lt m.
• A table entry contains both the node identifier
  number and the IP address of the node.

14
Characteristics of Node Table

• Each node stores information about only a small
  number of other nodes.
• A node has more information about the nodes
  closely following it on the identifier circle
  than about nodes farther away.
• A node table generally does not contain enough
  information to determine the successor of and
  arbitrary key.
• A Node can forward the request to another node in
  its table that has more information about a keys
  identifier.

15
Searching

• Node n searches its table for the node j whose
  identifier most immediately precedes key ks
  identifier.
• n passes the query onto j.
• Process is repeated until the node responsible
  for ks identifier is found.

16
Table for Node 1
Start 2 3 5
Int. 2,3 3,5 5,1
Succ. 3 3 0
6
1
0
7
1
2
6
3
5
4
2
17
Big Picture

• Cyber Entities (a key) will create an identifier
  number based on keywords or a unique Cyber Entity
  ID.
• Next, the cyber entitys location data, will be
  mapped to the proper BioNet platform (node).

18
Big Picture (continued)

• This location data is the IP address of the
  BioNet platform the CE is currently residing.
• If the CE migrates, O( log n ) messages are sent
  to update the platform that holds the CEs
  location data.

19
Other aspects to Consitent Hashing

• More complicated to explain (to prepare slides ?)
  is node joining and rejoining.
• Node failure, etc.

20
Future Work

• Analyze network traffic associated when many CEs
  are migrating.
• Analyze what happens when CE location data is
  lost, in a node failure.

21
0
7
1
2
6
3
5
4