Why Tapestry is not Pastry

1
Why Tapestry is not Pastry

• Presenter
• Arnold Pears
• CoRE Group, Uppsala University

2
Source Material

• This presentation is based on the following.
• Slides from the ROC/Sahara Retreat 2002
• http//www.cs.berkeley.edu/ravenben/tapestry/ROC-
  Sahara-1-2002.pps
• Tapestry Technical Report
• http//www.cs.berkeley.edu/ravenben/publications/
  CSD-01-1141.pdf
• Application Papers
• Oceanstore http//www.cs.berkeley.edu/ravenben/
  publications/pdf/asplos00.pdf
• Bayeux http//www.cs.berkeley.edu/ravenbe
  n/publications/pdf/bayeux.pdf
• Brocade http//www.cs.berkeley.edu/ravenbe
  n/publications/pdf/brocade.pdf

3
What is Tapestry?

• A prototype of a decentralized, fault-tolerant,
  adaptive overlay infrastructure(Zhao,
  Kubiatowicz, Joseph et al. 2000)
• Network substrate of OceanStore
• Routing Suffix-based hypercubeSimilar to
  Plaxton, Rajamaran, Richa (SPAA97)
• Decentralized locationVirtual hierarchy per
  object with cached location references
• Dynamic algorithms using local information
• Core API
• publishObject(ObjectID)
• routeMsgToObject(ObjectID)
• RouteMsgToNode(NodeID)

4
An Overview of Features

• A brief comparison of Pastry with Tapestry

5
Routing and Location

• Namespace (nodes and objects)
• 160 bits length ? 280 names before name collision
• Each object has its own hierarchy rooted at Root
• f (ObjectID) RootID, via a dynamic mapping
  function
• Suffix routing from A to B
• At hth hop, arrive at nearest node hop(h) such
  that
• hop(h) shares suffix with B of length h
  digits
• Example 5324 routes to 0629 as follows5324 ?
  2349 ? 1429 ? 7629 ? 0629
• Object location
• Root responsible for storing objects location
• Publish / search both route incrementally to root

6
(No Transcript)
7
(No Transcript)
8
Fault-tolerant Routing

• Strategy
• Detect failures via soft-state probe packets
• Route around problematic hop via backup pointers
• Redundant Access to Objects
• Via multiple root nodes for each object
• Object roots are well chosen within the net
• Protocols
• First Reachable Link Selection (FRLS)
• Proactive Duplicate Packet Routing

9
(No Transcript)
10
The Applications

• Tapestry
• OceanStore file sharing
• Bayeux scalable multicast for streaming
  applications
• Brocade Landmark routing
• Pastry
• PAST file sharing
• SCRIBE publish/subscribe service

11
Why is Tapestry not Pastry Then?

• Object location, replication and routing.
• Tapestry allocates root nodes to each content
  item and inserts hints into the network when
  the object or a replica register themselves
  with the root.
• ? multiple root nodes eliminate single point of
  failure
• Pastry creates a NodeID for each entity to be
  inserted into the overlay and inserts the
  resulting node into the network based on the
  NodeID, no hints or replication are managed
  by the substrate.

12
Why is Tapestry not Pastry Then?

• More on replication ......
• Pastry PAST replicates to the k nodes nearest
  to the original in the NodeID space. Content
  objects are replicated.
• Tapestry replicates hints along the "path from
  the server to the root". "Hints", i.e. references
  and metadata related to the original, are
  replicated.

13
Why is Tapestry not Pastry Then?

• Substrate knowledge of IP Net costs
• Proximity metrics relating the overlay to the
  underlying IP net topological costs are
  implemented slightly differently. One can argue
  that this is a cosmetic detail, and any arbitary
  cost strategy (such as location binningRatnasamy
  et al.) can be used in either system without any
  impact on the essential functionality.

14
Conclusions

• Tapestry and Pastry are extremely similar!
• Some confusion arises when comparing the two due
  to the division between the substrates and the
  file sharing applications built on them
  OceanStore and PAST respectively.
• Routing in both is based on Plaxton et al.
• Replication is handled in Tapestry via hints, but
  left to the application in Pastry
• Associating overlay routing with the underlying
  IP network structure via locality metrics is
  managed differently, but perhaps this is not
  significant.