Mesh-Based Content Routing using XML

1
Mesh-Based Content Routing using XML

• Alex C. Snoeren,
• Kenneth Conley, David K. Gifford
• MIT Laboratory for Computer Science

2
The Goal

• Reliably serve time-critical data streams
• To clients interested in different content
• Over a failure-prone and lossy Internet

3
Air Traffic Control Data

• Different client requests
• United Airlines flights
• Flights below 30,000 feet
• Takeoffs landings at BOS
• Time-critical data
• Lost or delayed data can be catastrophic
• Other domains have similar demands
• Sensor networks, live multimedia, etc.

4
Pub/Sub XML Style

• Servers generate an XML stream
• Annotate each XML packet individually
• Clients provide XML queries
• Parsing and query application is schema agnostic
• Network prunes content appropriately
• Internal nodes perform stream erasure to remove
  unwanted stream elements
• Application-level queries can be pushed into the
  network and combined

5
XML Encoding

• 153014022245CCZVTZ UAL1021 512 290 4928N12003W
• lt?xml version1.0?gt
• ltmessageidgt153014022245CCZVTZlt/messageidgt
• ltflightgt
• ltidgtUAL1021lt/idgt
• ltflightleg statusactivegt
• ltspeed type groundgt512lt/speedgt
• ltaltitude type reported mode
  plaingt290lt/altitudegt
• ltcoordinategtltlatgt4928Nlt/latgtltlongt12003Wlt/longtlt/c
  oordinategt
• lt/flightleggt
• lt/flightgt

Only United flights starts-with(string(/flight/id
),UA Flights below 30,000ft /flight/flightleg/a
ltitudelt300
6
XML Router Architecture

• Input DCP component
• Maintain parent set
• Assemble data stream
• XML Switch
• Parse incoming XML stream
• Route XML packets based on link queries
• Output DCP component
• Manage client subscriptions
• Package and distribute XML streams to clients

7
Coping with Failure

• Can handle node or link failure in two ways
• Component repair or replacement
• Takes time to detect failure and identify
  replacement
• Synchronizing replacement takes additional time
• Node and link redundancy
• Additional resources may need to be reserved
• Keeping standbys synchronized adds overhead
• Conclusion Use hot standbys, but
• Make effective use of existing resources
• Derive improved performance in the general case

8
Mesh-Based Routing
9
N-Resilient Mesh Construction

• Need to satisfy only three requirements
• Each node needs (n1) distinct parents
• Mesh must always be acyclic
• Paths to parents should be distinct
• Can be reconfigured on-line
• Nodes dynamically adjust their position
• By moving across or higher up
• Allows performance tuning
• Requirements may not be known a priori

10
Mesh Tuning
11
Dealing with Loss

• Can handle data loss in two ways
• Data retransmission (ARQ)
• Adds latency for NACK and subsequent
  retransmission
• Fact Internet latency is physically bounded from
  below
• Data redundancy (FEC)
• Reduces available bandwidth due to extra data
• Fact Number and capacity of Internet links is
  increasing
• Conclusion Send redundant data, but
• When it is likely to be most useful, and
• Where it will be most effective

12
Redundant coding

• Effectiveness of single-path FEC is limited
• Routing failures heal slowly Labovitz 00
• Packet loss is bursty Bolot 93, Paxson 99
• Use erasure coding across paths
• Erasure codes enable efficient encodings, but
• High efficiency requires large numbers of packets
• Decoding requiring multiple packets increases
  latency
• Carousel-based techniques add delay
• Duplicate packets can actually decrease latency
• But may not be as bandwidth efficient
• Hop-by-hop retransmission decreases recovery time

13
Diversity Control Protocol

• Combines data from multiple sources
• Packets given appl. seq. nos. (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

14
Emperical Loss Independence
0.1
Dispersity Routing
(Bottlenecked Hosts)
Back-to-Back
0.08
0.06
Effective loss rate
0.04
0.02
0
0
0.02
0.04
0.06
0.08
0.1
Observed packet loss rate
15
Reduced Latency
16
Summary

• Content-Based XML streaming
• Each packet is its own XML document
• Network is schema agnostic
• Diversity Control Protocol
• Reassembles stream from multiple sources
• Allows stream erasure while detecting loss

• Mesh-based overlay routing
• Provides tunable resilience to failure loss
• Can improve performance in general case