Tapestry: A Resilient Global-Scale Overlay for Service Deployment

1
Tapestry A Resilient Global-Scale Overlay for
Service Deployment

• Ben Y. Zhao
• IEEE JOURNAL ON SELECTION AREAS IN
  COMMUNICATIONS, VOL.22, NO. 1, JANUARY 2004

2
Introduction of Tapestry

• Tapestry is a peer-to-peer overlay network that
  provides
• High performance
• Scalable
• Location-independent routing of messages
• Using only local resource

3
Objective of this paper

• Provide decentralized object location and routing
  (DOLR)
• Message routing
• Object Replication
• Fault Tolerance
• Object Placement

4
Tapestry Routing Mesh
Nid
Outgoing Link
5
Path of a Message
Aid or GUID OG (42AX)
GR
6
Tapestry Object Publish Example
GR
GUID
7
Tapestry Route to Object
8
Message Processing
9
Tapestry Component Architecture
IP Address, Port, Nid
TCP/UDP
10
DOLR Network API

• PublishObject (OG, Aid)
• Best-effort, Receivers no confirmation.
• UnpublishObject (OG, Aid)
• Best-effort.
• RouteToObject (OG, Aid)
• Route message to location of an object with GUID
  OG.
• RouteToNode (N, Aid, Exact)
• Route message to application Aid on node N.

11
Tapestry Upcall Interface

• Delivery(G, Aid, Msg)
• Forward(G, Aid, Msg)
• Route(G, Aid, Msg, NextHopNode)

12
Node Insertion

1. Need-to-know nodes are notified of N.
2. N might become the new object root.
3. Construct routing table for N.
4. Nodes near N are notified and may consider using
   N in their routing tables.

13
Node Delete

• Voluntary Node Delete
• Notify the set D of Ns backpointers.
• Join republish traffic to N its replacements.
• Involuntary Node Delete
• Send beacons to detect outgoing link and node
  faulures.
• Build redundancy into routing tables.

14
Tapestry Implementation
15
Evaluation

• We post our wild-area experiments on PlanetLab, a
  network testbed consisting of roughly 100
  machines at institutions in North America,
  Europe, Asia, and Australia.
• In large-scale topology, we perform experiments
  using the Simple OceanStore Simulator(SOSS).

16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
400 Nodes, 62 PlanetLab Machines
All pairs round-trip routing latency (ms)
Percentile(Array, K)?????,???K???????
20
Relative delay penalty (RDP), Min distance
between the source and that endpoint
21
K backup node of the next hop of the publish
path, the nearest l neighbors of the current
hop.(1090 Nodes)
Relative delay penalty (RDP), Min distance
between the source and that endpoint
22
Insert a single node 20 times
23
Insert a single node 20 times
24
Delete a single node 20 times
25
MAX 1000 nodes
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)