Delay Tolerant Networks

Kevin Fall, "A Delay-Tolerant Network Architecture for Challenged Internets"

This paper is abstract. It suggests a overlay network which is running on networks different from the usual Internet and hence providing a very different service model. But there is no implementation (except one mentioned in one sentence in the conclusion) and no results section, no new algorithms.
I wonder whether the picture in Figure 1 really is a typical example: are there really applications where challenged networks are used for transit, or aren't most applications (e.g. sensor networks, satellite connections, ...) just having challenged networks at one or both ends?
And regarding the US Postal System as model for challenged Internets: looking back at my experiences so far, I kind of doubt whether this is a good idea...

Targeted networks

• Terrestial Mobile Networks
  
• Excotic Media Networks (e.g. deep space, acoustic links)
  
• military Ad-Hoc Networks
• Sensor Networks

Violated Assumptions

• E2E path exist, no disconnection
  
• max RTT not excessive, reasonable queuing time
  
• drop probability <<>

Possibilities to deal with the violated assumptions

• fix link-layer
• attach challenged networks at the edge of Internet through proxy, e.g. sensor networks, protocol boosters (e.g. HTTP proxy for satellite connections)
• electronic mail (mostly reliable, likely failure notification)
• DTN
  

New principles of DTN

• Message-switching (instead of packets)
• Regions with DTN gateways inbetween
• name includes region and entity name within that region
• several services (motivated by USPS)
• path selection and scheduling, trying to predict when the next contact to a certain region takes place
• custody transfer and reliability, e.g. if a message is sent from one node to another node, does the old node need to keep a copy until confirmation the message was delivered
• convergence layers
• time synchronization
• congestion and flow control