Transcript Paper Presentation: "A Delay Tolerant Network Architecture for

A Delay-Tolerant Network
Architecture for Challenged
Internets
Author: Kevin Fall
Paper Presentation: Vinay Goel
Internet Service Model
 TCP/IP based
 End to end IPC using concatenation of
potentially dissimilar link layer
technologies
 Packet-switched model of service
 A number of key assumptions…
Key Assumptions
 Assumptions
 End to end path exists b/w data source and its
peer(s)
 Maximum RTT b/w any node pairs is not
excessive
 End to end packet drop probability is low
 A class of challenged networks violate one or
more of the assumptions
Examples of challenged
networks
 Terrestrial Mobile Networks
 Unexpectedly partitioned due to node mobility,
changes in signal strength etc.
 Exotic Media Networks
 High latencies with predictable interruption,
suffer outage due to environmental conditions
etc.
 Military Ad-hoc Networks
 Hostile environments
 Mobility, environmental factors, or intentional
jamming may cause disconnection
 Data traffic competing for bandwidth
Characteristics of these
networks
 Path and Link characteristics
 High latency, low data rate
 Disconnection
 Long queuing times
 Network Architectures
 Interoperability considerations
 Security
 End system characteristics
 Limited longevity
 Low duty cycle operation
 Limited resources
Adapt Internet to these
environments?
 Link-repair approaches
 Engineer problem links to appear more similar to
the types of links for which TCP/IP was designed
 “fool” the internet protocols: strive to maintain
end-to-end reliability etc.
 Attach these networks to the edge of the
Internet
 Use of a special proxy agent
 Provides access to and from challenged networks
from the Internet
 No support for using such networks for data
transit
Link repair approaches
 In-network entities (“middle boxes”)
 Performance Enhancing Proxies (PEPs) &
protocol boosters
 Contain state necessary for connection violating
the Internet fate sharing principles
 Confound end-to-end diagnostics and reliability,
increase system complexity if mobility is frequent
 Pose a significant challenge for end-to-end security
mechanisms
Application Layer proxies
 Provide specialized Internet-to-”special
network” name mapping & protocol
translation
 Used at the edge of special networks
 Disadvantage: their specificity
 Either respond to a specialized set of commands
or act as raw data conduits
 Limit the ability to re-use proxies for different
applications
 Fail to take advantage of special resources (storage,
processing capabilities etc.)
Electronic Mail
 Asynchronous message delivery system
 Provides an abstraction that comes close to
addressing many problems
 Flexible naming, asynchronous message-based
operation etc.
 Falls short
 Lack of dynamic routing
 Weakly defined delivery semantics
 lack of consistent API
What’s the most desirable
framework?
 A network service and API providing
non-interactive messaging
 System should combine some overlay
routing capability (such as in P2P
systems) with delay-tolerant and
disconnection-tolerant properties of email
Delay Tolerant Message Based
Overlay Architecture
 Based on abstraction of message switching
 Message aggregates known as “bundles”
 Routers that handle them are called “bundle
forwarders” or DTN gateways
 Architecture provides a store-and-forward
gateway function between various network
architectures
Regions and DTN gateways
 Two nodes are in the same region if they can
communicate without using DTN gateways
 DTN gateway
 Point through which data must pass in order to
gain entry to a region
 Can serve as a basis for both translation and well
as a point to enforce policy and control
Name Tuples
 Identifiers for objects or groups of objects
 DTN name tuple {Region Name, Entity Name}
 First portion is a globally unique,
hierarchically structured region name
 Interpreted by DTN gateways to find the path(s)
to one or more DTN gateways at the edge of the
specified region
 Second portion identifies a name resolvable
within the specified region
 Need not be unique outside the region
Name resolution
 Only region identifier is used for
routing a message that is in transit
across a collection of regions
 Entity name information is locally
interpreted in the destination region
 Form of late binding
A Postal Class of Service
 Priority based resource allocation
 Adopt a subset of the types of services
provided by US Postal Service
 Attractive characteristics
Low, ordinary and high priority delivery
Return receipt, delivery records
Path Selection and Scheduling
 Architecture targeted at networks where an
end-to-end path can’t be assumed to exist
 Routes are comprised of a cascade of timedependent contacts (communication
opportunities)
 Particular details of path selection and
scheduling - heavily influenced by regionspecific routing protocols and algorithms
Custody Transfer and
Reliability
 Custody transfer: acknowledged delivery of a
message from one DTN hop to the next and
corresponding passing of reliable delivery
responsibility.
 End hosts do not ordinarily need to keep a
copy of data that has been custodially
transferred to a DTN next hop
 Custody transfer can be viewed as a
performance optimization for end-to-end
reliability that involves endpoint movement
Convergence Layers and
Retransmission
 Facilities provided by transport protocols in
use within the regions may vary significantly
 Bundle forwarding assumes underlying
reliable delivery capability with message
boundaries when performing custody
transfer
 Transport protocols lacking these features must
be augmented
 Include transport-protocol-specific
convergence layers
Time Synchronization
 Coarse Level
Identifying message fragments
Purging messages that have exceeded
their source specified lifetimes
 Stringent constraints
Scheduling, path selection
Congestion management
Security
 Verifiable access to the carriage of traffic at
a particular class of service
 Avoid carrying traffic potentially long
distances that is later found to be prohibited
 Each message includes an immutable
“postage stamp” containing
 Verifiable identity of sender, an approval, class of
service etc.
 Credentials checked at each DTN hop by
routers; use of public key cryptography
Congestion and Flow Control
 Flow control: limiting the sending rate
of a DTN node to its next (DTN) hop
Attempt to take advantage of underlying
protocols’ mechanisms
 Congestion control: handling of
contention for the persistent storage of
a DTN gateway
Shared priority queue for allocating
custody storage
Application Interface
 Applications must be careful not to expect
timely responses
 Must be capable of operating in a region
where a request/response RTT may exceed
the longevity of the client and server
processes
 Structured to continue operating in the face
of reboots or network partitioning as much
as possible
Conclusion
 Design embraces notion of message
switching with in-network storage &
retransmission, late-binding of names
& routing tolerant of network
partitions
 Puts forth several design decisions
worthy of consideration
Questions?