Meshbased Content Routing Using XML

1
Mesh-based Content Routing Using XML

• Authors Alex C. Snoeren (MIT)
• Kenneth Conley (MIT)
• David K. Gifford (MIT)
• Presented By Moosa Muhammad
• (some slides borrowed from authors presentation)

2
Introduction

• A new approach for reliably multicasting
  time-critical data to clients, over mesh-based
  overlay networks.
• Primary motivation is to deliver information to
  end-clients with low latency in the presence of
  both node and link failures.
• In real-time applications, a small delay in a
  data feed may be unacceptable.

3
Introduction (cont.)

• Past work in overlay networks has shown that
  multiple, distinct paths often exist between
  hosts on the Internet.
• Basic approach is to construct a content
  distribution mesh, where each node is connected
  to n parents, receiving duplicate packet streams
  from each of its parents.
• By maintaining an acyclic mesh, this approach
  guarantees an (n-1)-resilient mesh, without
  repair.

4
A Mesh Network
5
Publish/Subscribe model

• Data streams consist of a sequence of XML packets
  and are forwarded by application-level XML
  routers.
• The routers use a novel Diversity Control
  Protocol (DCP) for router-to-router and
  router-to-client communication.
• Clients subscribe to content by providing a
  query that describes the portions of the XML
  stream they would like to receive.

6
Resilient Mesh Networks

• Clients wishing to join an (n-1)-resilient mesh
  perform the following steps
• Compose XML query
• Contact n existing routers that can service the
  query
• Send these n routers the XML query
• Receive the XML stream described by the query
• Each router maintains a query table that
  describes the portion of the XML each of its
  children wishes to receive.

7
XML Routers

• XML Router Core
• Evaluate each received XML packet against all
  output link queries
• Diversity Control Protocol
• Implements resilient mesh communication by
  allowing a receiver to reassemble a packet stream
  from diverse sources
• Mesh Initialization and Maintenance
• A set of algorithms to automatically organize
  routers and clients into a mesh and repairs the
  mesh (when faults occur)

8
Diversity Control Protocol

• Combines data from multiple sources
• Packets given application sequence s (ANs) by
  root nodes
• Uses first copy of each packet received
• Detects packet loss
• Each packet includes previous packet AN
• NACK-based retransmits
• Enables stream erasure
• Modify previous AN field

9
DCP (cont.)

• For a given content stream, packet identifiers
  are associated only with the packet content and
  are not sender specific.
• Packets are buffered and transmitted in-order at
  each hop.
• DCP currently uses UDP as a transport mechanism.
• One DCP packet is used to transport one XML
  packet.

10
Mesh Formation and Maintenance

• Parent selection algorithm
• Initialize the set S to be the root routers
• For each node in S, send a join request and
  remove the node from S
• If a node accepts the join, add it to the parent
  set P. If n nodes are in P, quit
• If a node declines the join, ask it for a list of
  its children, and add them to S
• If S ! empty, jump to step 2
• The method to obtain a new parent (i.e. mesh
  repair) is currently identical to the above, with
  one caveat.

11
Experimental Design

• Results presented in this section are an average
  of several experiments each consisting of 1000 to
  10,000 XML-encoded packets.
• The roots and all intermediate router nodes were
  run on one machine. The XML client node was run
  on another machine.
• The desired link loss rates between intermediate
  and client nodes were obtained by routing each
  DCP connection through another machine which
  passed the packets through a Dummynet tunnel.

12
Client v.s. Parent Loss Rates
13
Throughput v.s. Loss Rate
14
DCP Performance Summary

• DCP has several attractive features independent
  of the format of the data stream.
• DCP-based meshes can achieve substantially lower
  effective loss rates and latency than tree-based
  distribution networks.
• Redundancy can be utilized to absorb unexpected
  decreases in link capacity between nodes.

15
Air Traffic Control Data

• Original motivation was to build an
  infrastructure for distributing and processing
  real-time air traffic control data.
• Converted ASDI feed into XML packet stream.
• Simply converting the feed resulted in
  approximately four-fold increase in bandwidth
  consumption.
• Fixed this by applying Lempel-Ziv data
  compressor.
• Since application serial numbers (ANs) were
  provided by the FAA, synchronizing multiple roots
  was straightforward.

16
Conclusion and Future Work

• Three key ideas presented
• XML routers that switch self-describing XML
  packets
• Showed how XML routers can be organized into a
  resilient overlay network that can tolerate both
  link and node failures
• Introduced Diversity Control Protocol as a way to
  reduce latency and improve reliability
• Future work can include extending XML routers to
  support duplex communication and coming up with
  more sophisticated XML mesh building
  maintenance algorithms.