Title: DTN GoNoGo Phase I to II
1Disruption Tolerant Networking (DTN) Program
Phase 1 And Phase 2/3 Program 23 March 2006
Preston Marshal, Program Manager Preston.marshall_at_
darpa.mil 703-696-5273
2Disruption Tolerant Networking (DTN)Program
Concept
End-to-end severely disrupted by one bad link
Custodian-to-custodian connections isolate
disrupted regions
DTNs Goal is to is to develop and demonstrate
technology that will provide network services in
the face of disruption and massive differences in
delay and bandwidth and to reduce demands on
network resources by integrating storage into the
network
3Military Need
FCS Vehicle, Ft. Benning 2006
- FCS Communications Position Reports Used as
Measure - Highly Favorable Metric Used
- Loss of 2 Successive (1 Sec Interval) Reports
Considered as Disconnected
Wireless networks can be good for local connect,
but often cant reach back to infrastructure Local
storage caching can create access to
information after infrastructure connectivity
loss.
- Relying on IP for tactical military networks is
dangerous - Episodically connected military MANETs see rapid
topology changes - Tactical radios know names, not destination
addresses - Tactical/edge military networks may be a mix of
IP and non-IP radios
4All Bandwidth is Not Equally Important
- Networks are not hierarchies of bandwidth, they
are islands - Bandwidth within islands not as important as
bandwidth between islands - DTN augmentation within islands provides major
performance benefits between islands - Nodes can use local bandwidth to obtain DTN
services, even if not on own node
- Similar to DARPA
- Highly reliable, high speed (1 Gigabit) from
servers on campus - Several Megabits in and out to Internet
Bandwidth/Reliability
64 Kilobps Episodic Connectivity
GIG Fiber Core
Distance
10 Gigabps Highly Reliable Connectivity
DTN Can Augment Existing Networks without Being
Inserted into Topology
10 Megabps Highly Reliable Connectivity
Wireless Enclaves
5DTN Network Persistence Can Solve Fundamental
Internet Application Shortfalls
- DTN makes applications over disrupted networks
robust - DTN is also an Opportunity to solve Fundamental
Problems weve never before had a handle on,
using Network-Managed Persistence - Access information by content or type rather than
by network address - I want maps for my area instead of I want to
ftp to 192.168.4.17 - Retrieve once, provide to local users as
requested - Learn from actual network usage
- Exploit in-network storage/caches and pub/sub
protocols to create a dynamic and self-forming
Akamai - Use temporal security rather than physical
security
Temporal Security Model
Time
6Todays Network Push or Pull, Neither Optimal
Conventional Pull Copies to every requestor
Only those who ask, get but with delay N
requests use N times the bandwidth
Multicast Push Data goes to everyone
I need a map
I need a map
I need a map
I need a map
I need a map
Connected Network
I need a map
Connected Network
I need a map
I need a map
I need a map
I need a map
I need a map
Only one transfer, but data flows to everyone in
the multicast group, not necessarily when / where
the data is needed
I need a map
DTN Resolves Both Inefficiencies.. Pulls One
Time, Distributes Locally To Requestors
7DTN Phase 1 Results
- Demonstrated DTN v TCP with typical USMC wireless
connectivity patterns (MITRE/CONDOR) - Demonstrated Network Delivery (BBN)
- Demonstrated Trusted Delivery Resistance to
DDoS (Lehigh) - Designed architecture intrinsic ability of DTN
to operate to the extremes of the network without
segmenting to match network characteristics
meta-architecture (MITRE/JPL) - Potential to move this extensible framework to
other building blocks of the network - Have to adapt Cisco/Nortel/Lucent/Juniper
behaviors - Implemented Experimental Operating Wireless DTN
(GaTech/UMass)
8Demonstrated DTN v. TCP with USMC Wireless
Connectivity Patterns
INMARSAT terminal
Cisco 2811
KG-250
DTN
10 KByte File Transfers in 24 hours
Abandoned 10-KByte File Transfers in 24 hours
4000
140
user
Cisco 3725
EPLRS
3500
120
HTTP
HTTP
3000
CONDOR Gateway cable map
DTN
100
DTN
2500
80
2000
3580
..
DTN Is A Deployable Technology With Massive
Performance Benefits for DoD
115
60
1500
1000
40
500
20
368
0
0
0
Completed
Abandoned
Demonstrated that DTN is Useful Feasible, and
that DTN can be Transitioned to COTS-based
Military Systems
9Phase 1 Go/NoGo Metric Demonstrate DTN Network
Performance in Disrupted Network Evaluation
Platform Hardware in the loop emulation of
actual DTN nodes
- 100 Reliable Delivery with
- 80 Utilization over
- 20 Available Links
ACHIEVED
- Link characteristics
- capacity 19.2 kb/s
- delay 5 ms
- MTU 1480 bytes
- Bundle traffic
- size 2800 bytes
- total originated 264
- Network Transit
time gt620ms - Link StateTransit time
4.3s - Mean time between link transitions 5s
- Run time 3600 s
DTN would have delivered all traffic given enough
time
For random link dynamics, at most 16 (out of 31)
bi-directional links were up at any time
Network changes faster than it updates.. never
static. IP would never have correct topology..
would fail in a conventional MANET
10Delivered Bundles Vs. Path DistanceRun at 20
Target Availability Random Link Dynamics
- Opportunistic Routing Found Ways to Deliver All
Traffic, Regardless of Hops - TCP (End to End) Could Not Find Opportunities
- End to End requires Complete Path be Available
- End to End is Fundamentally Unsuited for Military
Operations - 80 Links are only 20 Network Connected at 7
Hops - 20 Links are 0.001 Network Connected at 7 Hops
- End to End IP (Without TCP) Shares All these
Issues
Delivery Performance for DTN and TCP
DTN
TCP End to End Transfer
11Delivery Ratio Worst Case DynamicsDTN versus
End-to-End (E2E) Baseline
- DTN Accomplished All Deliveries for
Availabilities Above Go/NoGo Criteria - Would Complete All if Longer Duration created
Opportunities - End to End Could Not Find Sufficient
Opportunities in Any Disrupted Scenario - Failed Completely Below 50 Availability
DTN
End to End
12Link Utilization Using DTN
- DTN Effectively Used All Available Link Capacity
- Network Was So Dynamic that End to End Would not
be Aware of Opportunities to Use - Efficiency Decreases at High Availability, as
More Overhead, and Early Completion of Transfers - Phase 2 Will Develop Technology to Adapt and Use
both End to End and DTN Based on Which Would be
Most Effective
13Trusted Delivery GNG Metric ACHIEVED
Phase 1 Go/No Go Demonstrate Trusted Delivery
?
- Demonstrate rejection of message from
unauthenticated sender - Demonstrate authentication and forwarding of
message from trusted sender - Demonstrate payload data encryption
?
?
DTN will not propagate Distributed
Denial-of-Service Attack DTN will Detect Reject
Fraudulent (Forged Address) Messages
14Trusted Delivery GNG Metric ACHIEVED
Demonstrate rejection of message from
unauthenticated sender
- Two sending nodes - one legitimate, one malicious
- attempt to send a bundle in a network with the
BAH feature enabled - The malicious node (M1) sends a bundle without
the appropriate BAH to the forwarding node (N2) - Result N2 rejects the bundle - ACHIEVED
- The legitimate sender (N1) sends a bundle with
the appropriate BAH, allowing for successful
authentication - Result N2 forwards the bundle to the
destination (N3)
BAH Bundle Authentication Header
Security Perimeter
N1
N2
N3
M1
Should have been part of the Internet from the
beginning
15Trusted Delivery GNG Metric ACHIEVED
Demonstrate 1.) Authentication and Forwarding
of Message From Trusted Sender and 2.) Payload
Data Encryption
- N1 sends a bundle to N4 (thru N2) with only the
BAH activated - The link between N2 and N3 is insecure, so policy
at N2 requires payload data encryption - N2 encrypts the payload, adds the PSH, and
becomes the PSH-source, with destination N4 the
PSH-destination for the bundle - N4 receives the encrypted bundle from N3 (thru
N2) and decrypts the message ACHIEVED
N3
N1
N2
N4
PSH-Source
PSH-Destination
BAH Bundle Authentication Header PSH Payload
Security Header Red Cleartext Black Ciphertext
DTN Enables Security Partitioning Based on
Traffic Policies Rather than Physical Topology
16DTN System Architecture
Bundle Engine
API Legend
Protocol Composition API
Management API
Autoconfiguration/ Neighbor Discovery
Bundle Custody Transfer
Bundle Encryption
Bundle End-to-end Reliability
Bundle Flow/Congestion Ctl.
Bundle TBD Services
DTN Policy/Management
DARPA Routing Protocol
Environmental Awareness
DTNRG Routing Protocol
Other Routing Protocol
RoutingAPI
Configuration API
EnvironmentalAwareness API
Convergence Layer
Plug-ins/DLLs
Process Rendezvous
Single DTN Standard Will Be Extensible for
Commercial or Uniquely Structured Military Apps
Such As UAV Overflight, Sensor Nets, Tactical
Disruption
17Technology for a Common Routing Structure with
Mission-Unique Algorithms
- Wireless networks need diverse routing behaviors
- Open Biggest Battery First (Battery-powered
systems) - Use Advantaged Node Last (Transient aircraft
nodes) - Open Least Tx Energy Path First (Energy-starved
systems) - Open Least Used Reasonable Path First
(Fairness) - Extend - dont replace - COTS products
Commercial World
DoD Infrastructure
DoD Sensor Field
minimal protocol set
Core/Interoperable
Core/Interoperable
Core/Interoperable
GIG-unique routing algo.
battery-aware routing algo.
UAV flight schedule
UAV flight schedule
Core/Interoperable
vendor-unique extension
Buy commercial, specialize to military
Color Legend
Commercial DTN Extension
IRG DTN Network Standard Core
Military DTN Extensions
18DieselNetInitial Deployment May 2004
- University DTN testbeds (GaTech/UMass) urban ops
experiment with multipath and rapid topology
change (route breakage) - Long-term 24/7 Experiment at Low Cost with Mobile
nodes, sensors, and throwboxes analogs of
tactical military wireless networks urbanrural
manned vehicular
DieselNet routers in 40 busses in Amherst
19Algorithmic Results
- Knowledge management Uniform information
dissem-ination and improvement of buffer usage - Resource management Virtual infrastructure with
transport frames improves delivery rate in
bottleneck scenarios - Opportunistic Routing SCaTR framework improves
delivery rate and reduces signaling overhead - Reflective Route Planning First DTN routing
algorithm based on formal reasoning technology - Flexible network simulation models with
user-defined physical resource schedules
20DTN Progressive Maturity
- Integrate Push and Pull Metaphors
- Cognitive Caching
- Information Addressing (not Network Addressing)
- Multiple Native Networks (JTIDS, IP, EPLRS, )
- Initial Demo Board Implementation
Protected, High Performance DTN for Static
Applications with Store and Forward
Phase 1
- Self-Organizing in Response to Network needs
- Large Scale
- Red/Black Management of Persistent Data
Phase 1 Protected, DTN for Medium Scale, Static
Applications with Caching and Distributed Query
Phase 2
Dynamically Self-Organized Organized, Secure
Local Store, Application Linkages, Proven
- Demo in Military Scenario to Assess Utility
- Implement in Longer term, non-Military
Application for Operational Experience
Progressive Technology Development Resulting in
Proven and Deployable Product
Phase 3
- Integrate into Military Networks
- Implement in Longer term non-Military Application
to Acquire Experience
21Merging Information and Networking
- Policy reasoning enable sophisticated queries
over the network - I dont know exactly what Im looking for, but I
know how to describe it - Late binding as a way of describing information
- Dont have to know where information resides
Google as a metaphor, not an overlay - Late binding can occur in the information domain,
not only the addressing domain - Want to build a formal structure for persistence
and networking, a structure for solving tactical
problems - Analogous to akamai, but akamai is static.. In
tactical networks must build our persistence
architecture on the fly
22Adaptation to Reflect Network Dynamics
- DTN networks adapt to changing network topologies
- Storage configures itself around paths thru the
(intermittent) network - Self-forming Akamais for content distribution in
response to network demands - Caching as a result of delay-bandwidth product
discontinuities - Military Utility Reduce (eliminate?) burden of
planning network deployment with unit deployment - Planning costs currently comparable to or greater
than people and equipment costs - Network planning creates inertia/delay in
deploying forces and reacting to unanticipated
changes in the theatre - Avoid the Comms planning cycle!
23Content-based Networking
- Support push from core, pull from edge, and
meet-in-the-middle content-based networking - Steinbet Users will pull data as needed instead
of having massive amounts of information pushed
to them regularly regardless of whether it is
needed. .. a key tenet of net-centric warfare is
that the consumers of information are smarter
than their sources about what is needed
operationally right now, and that they should be
able to pull those data when they need it. - Enable users to subscribe to or query useful
information services, and have data returned when
theres a new event or query match - Edge networks can push data up into the network
- Source analysis systems can query DTN storage for
Wolfpack systems enables heterogeneous sensor
data fusion - Distribute policies with bundles much of the
flexibility of Active Networks without as much
risk .. Update rules of engagement by
disseminating policies thru DTN nets
24Benefits of unifying networking and storage
- Request information by content/type rather than
by network address - I want weather for my area instead of I want
to ftp to 192.168.4.17 - Ability to cache rather than waste wireless
bandwidth - Its way cheaper to store data rather than to
transmit it again - Integrating push-pull metaphor
- Pushing sends to everyone and wastes bandwidth,
can pre-place data - Pulling serves a single user, same data requested
multiple times wastes bandwidth, incurs large
delays delays in disrupted networks - Akamai uses static caches in a wired network to
mitigate bandwidth wastage and delay - DTN Push/Pull exploits DTN in-network storage
(persistent caches) and pub/sub protocols to
create a dynamic and self-forming akamai - Temporal security
- Show the data as encrypted/unencryptd
25A New Security Model
- Red-black separation derives from the philosophy
that the control center is protected once in
the black, info is physically safe - With low-cost devices like WNaN, no longer true
- How to deal with the loss of equipment at the
tactical edge? - Information on this equipment is compromised with
the equipment - How to change the security model to deal with
equipment that cant be physically secured?? - Rather than view red-black as physical
separation, think temporal separation! - Keep data encrypted unless the application is
processing it! - Encrypted data lives in local cache or edge
network cache, decrypted by app - Use a DTN security convergence layer shim for
apps .. Withdraw access by app by revoking cert
or similar action. - DTN mechanism protects information keyboard to
eyeball - Protection from app to app, not from node to node
Temporal Security Model
Time
26Summary
- Bigger Challenge!
- Larger Funding!
- Massive Need!