Transcript BBN_Krishnan

Krishnan
SPINDLEII Project Overview
Disruption-Tolerant Networking R&D at BBN
Rajesh Krishnan
[email protected]
On behalf of the BBN SPINDLE-II Project Team
Presented at the DTN Phase 2 Kickoff Meeting
August 9, 2006
DISTRIBUTION STATEMENT C
Distribution authorized to U.S. Government Agencies and their contractors (Critical Technology) (August 9, 2006).
Other requests for this document shall be referred to DARPA Technical Information Office.
© 2006 BBN Technologies
Krishnan
Outline
• BBN’s SPINDLE project
• A declarative knowledge-based approach to DTN
• Late binding of intentional names for endpoints
• Disruption-tolerant access to content
• Other research thrusts: routing and policy support
• Schedule and team
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
2
© 2006 BBN Technologies
Krishnan
BBN SPINDLE* Project
•
SPINDLE (DARPA DTN Phase 1)
Feb 2005 to Aug 2006
– routing approaches for DTN, and optimization via machine learning
• academic partner: University of Missouri, Rolla (reinforcement learning)
– policy-based resource management
– late binding of intentional name attributes
•
SPINDLE-II (DARPA DTN Phase 2)
Jul 2006 to Jan 2008
– robust, modular system based on open DTNRG standards and software
• enable DoD-relevant technology insertion via plug-ins developed independently
• develop specifications, e.g., for late binding, policy-based resource management
– vehicular DTN system prototype
• 70 units planned, to be available to DTN participants and DoD partners (via DARPA)
– innovative DTN technology development
• content-based networking, routing, adaptive dissemination, late binding, and policy
– demonstration in military-relevant scenario TBD
• seeking DoD partners who are willing to evaluate DTN by end of Phase 2
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
* Survivable Policy-Influenced Networking: Disruption-tolerance through Learning and Evolution
3
© 2006 BBN Technologies
Krishnan
System Research and Development
DTN System Software For GIG Insertion
DoD applications:
sensor data exfiltration, caching
common operational and tactical picture
Enterprise applications:
web services, database transactions,
middleware, filesystems
Basic applications:
email, chat, offline browsing, file transfer
Adaptive
Dissemination
Application
interface
System-wide access
to knowledge base
Storage/KB
interface
Persistent
Knowledge
Base
DTNRG Core
Platform
Policy
Engine
Decision plane
interface
Routing
Routing
Routing
Strategies
Strategies
Strategies
Convergence
layer interface
Late Binding
of Name Attributes
DoD links/networks
TCP/IP
Other COTS
• Robust modular DTN system software
- core platform: open source and standards
- plug-ins for DoD technology insertion
• e.g., convergence layers, storage/KB, routing, naming
• independently developed by DTN performers (including BBN), or others
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
4
© 2006 BBN Technologies
Krishnan
System Research and Development
Vehicular DTN System Prototype (Notional)
Example: Parvus DuraCOR-810
•
Example: Tri-M CANTAINER
Rugged PC-104 hardware with DTN system software installed
-
specifications to be determined in consultation with key DTN stakeholders
extended temperature range
weather-proof (NEMA compliant)
modular: various depths and faceplates
wide range of add-ons available for this form factor:
• I/O, communications, SMPS, batteries, ...
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
5
© 2006 BBN Technologies
Krishnan
Outline
• BBN’s SPINDLE project
• A declarative knowledge-based approach to DTN
• Late binding of intentional names for endpoints
• Disruption-tolerant access to content
• Other research thrusts: routing and policy support
• Schedule and team
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
6
© 2006 BBN Technologies
Krishnan
A Declarative KnowledgeBased DTN
• Knowledge bases (KB) offer a flexible and extensible framework
–
–
–
–
derivation of facts from simpler facts by applying inference rules
uniform query interface for both explicitly-stated and derived facts
decision making and search in a rich space of dynamic facts/rules
rapid prototyping and deployment of network capabilities
• Persistent knowledge base integrated into Phase 1 system
– object-oriented logic for knowledge representation (Flora-2)
– built on top of a Tabled Prolog system (XSB)
– backend persistence based on RDBMS (MySQL)
• Integrated declarative approach to routing, naming, and policy
– approach extends naturally to content-based networking
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
7
© 2006 BBN Technologies
Krishnan
Specifying Rich DTN Information
Example: Inference Rules For Adjacency Formation
t = T1
t = T2
Ground node X can form a Predicted Adjacency
with some Uav node at t=Tnow if trajectory
information is known beforehand. Useful since
Uav in range for a short duration, not enough for
traditional network topology discovery.
Uav
t = Tnow
X
•
Rule expression in KB to deduce a predicted adjacency
predictedAdjacency :: spindleAdjacency.
S : predictedAdjacency [from -> X, to -> Uav, upAt -> T1, downAt -> T2] :Uav [trajectory -> Trj1], X [trajectory -> Trj2],
walltime (Tnow), trajectory_crossing (Tnow, Trj1, Trj2, [ T1, T2 ]).
•
Contents of the KB can be disseminated for informed routing decisions
– forwarding, routing, and other decisions based on queries into the KB
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
8
© 2006 BBN Technologies
Krishnan
Outline
• BBN’s SPINDLE project
• A declarative knowledge-based approach to DTN
• Late binding of intentional names for endpoints
• Disruption-tolerant access to content
• Other research thrusts: routing and policy support
• Schedule and team
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
9
© 2006 BBN Technologies
Krishnan
Late Binding Of Name Attributes
•
Example: coalition force member addresses a bundle to
– target intentional name: teamLeader@{org=US Army,loc=36N,43E}
– not aware of canonical name: [email protected]
•
Name resolution often needs to be performed “late” in a DTN
•
Late binding services:
– not at the originator (e.g., unlike DNS lookups at source)
– but progressively within the network
– mapping of names from a rich namespace to routable endpoints
• extensible ontology to express multi-attributed intentional names
– e.g., based on physical location and trajectory (UAV flight plans)
• protocols to maintain name KBs and progressive resolution within network
– deferred binding of next hop to convergence layer address and parameters
• capability for several communication modes (e.g., WiFi, GPRS, ...) may exist
• parameters (address, protocol, data rate) may need negotiation upon contact
• enables routing with only coarse grained information about future adjacencies
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
10
© 2006 BBN Technologies
Krishnan
Progressive Resolution of Attributes
Endpoint Names As Queries
snet:Sname
…
snet:Sname
name
record
LB
header
name
ontology
Sname : sensor [
cname  X,
value(Temp)  T,
isIn(‘DeathValley’)
], T>130.
may-bind varlist {?X,?T}
WHERE
snet :: sensor [
cname  string,
value(Senstype)  int,
isIn(Loc)  bool,
…
].
nexthop Cname = dtn:n2
snet:Sname
rewrite LB
header
…
Sname : sensor [
cname  X,
value(Temp)  T,
long  116°49′33″W,
lat  36°14′31″N
], T>130.
may-bind varlist {?X,?T}
…
…
Sname : sensor [
cname  X,
value(Temp)  T,
long  116°49′33″W,
lat  36°14′31″N
], T>130.
must-bind varlist {?X}
may-bind varlist {?T}
nexthop Cname = dtn:n5
dtn:n3
Death Valley
Sensor
Network
nexthop Cname = dtn:n4
dtn:n5
dtn:n1
August 9, 2006
dtn:n2
SPINDLE-II Overview -- DTN Phase 2 Kickoff
dtn:n4
11
© 2006 BBN Technologies
Krishnan
Outline
• BBN’s SPINDLE project
• A declarative knowledge-based approach to DTN
• Late binding of intentional names for endpoints
• Disruption-tolerant access to content
• Other research thrusts: routing and policy support
• Schedule and team
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
12
© 2006 BBN Technologies
Krishnan
DisruptionTolerant Content Access
Self-Organized Caching And Search Capability To Go
Domain Name System
Phase 1 Vision
Late Binding of
Intentional Names
Strong connectivity
assumed, non-semantic Disruption-tolerant access
names, resolution must to network nodes; data
will be eventually delivered
occur at source
Universal Resource Locator
Information identifiers
based on location
or topology (even for
cached content)
August 9, 2006
Multi-attribute
names, queries as
names, progressive
resolution in network
Concept and approach
developed in Phase 1
To be refined and
implemented in
Phase 2 system
Content-Based Universal
Resource Identifiers
Concept and approach to
Phase 2 Vision
Disruption-tolerant Information identifiers
access to information
based on content,
secure self-organized
content caching and retrieval
SPINDLE-II Overview -- DTN Phase 2 Kickoff
be developed in Phase 2
To be implemented
in Phase 3 system
13
© 2006 BBN Technologies
Krishnan
An Imaginary Tactical Scenario
Not An Always-Connected Broadband Utopia
•
A thought experiment …
•
SGT Fictitio’s team is under attack and is forced by the enemy to deviate
from their planned mission. He needs a detailed map of a section of
Anarchypolis annotated with IED incidents and cultural cues. He also
needs to send information to his command post regarding his team’s
situation. He is temporarily disconnected from the GIG. Members of the
team are outfitted with computing and network gear, but SGT Fictitio has
only intermittent connectivity to them (in particular to his sniper, PFC
Smith).
•
Adopting the agile distributed network-centric warfare paradigm, SGT
Ficititio’s team share information locally in order to self-synchronize.
Some of the information sharing is still done manually. Team members
cannot easily access content on the team network unless they know the
specific file-name/URL or the information is sent to them via chat/email.
•
SGT Fictitio and his team members did extensive research about the
mission (over and above the material received from higher command).
They saved only the results deemed relevant to the planned mission.
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
14
© 2006 BBN Technologies
Krishnan
Scenario Without DTN
• SGT Fictitio must wait for stable end-to-end paths
– content location (URL) if known (e.g., particular map server
and IED incident database server), and the domain name
server (DNS)
– else, he must access a search engine URL that provides
content URLs, then a caching server that provides cache
locations for content URLs, and DNS
• Unfortunately he never gets the content he needs,
despite occasional contact with team members who
in turn are intermittently connected to rest of the GIG
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
15
© 2006 BBN Technologies
Krishnan
Scenario With DTN
• SGT Fictitio’s map request and message to command
post gets picked up by his team members’ computers
and get delivered using DTN
– he gets a map from the map server eventually
– reinforcements are dispatched by the command post
– annotations arrive later than the reinforcements
• The fact that a relevant map was cached on PFC
Smith’s computer was never discovered
– it had not been indexed by the central search engine
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
16
© 2006 BBN Technologies
Krishnan
Scenario With ContentBased DTN
• SGT Ficititio’s map request and message to command
post are relayed using DTN as before, however:
– he receives the map and annotations locally at the next
intermittent contact with his team
– map and relevant annotations were automatically cached
and indexed on PFC Smith’s computer during research
– these were also kept up-to-date opportunistically
• His team is able to devise a better game plan
– they return to base even before reinforcements arrive
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
17
© 2006 BBN Technologies
Krishnan
Another Content-Based DTN Scenario
Common Operational Picture
Opportunistic
Request:
Need
viewContent
of
Content
Astarget!
Up-to-date As
Opportunistic
Content
Possible
(both east Delivery
and
westUnder
views)Disruption
Delivery
Content-Based
Addressing
W E
West View
W
Opportunistic Content
Caching/Replication
E
East View
E
West View of Target
Generated/Stored at Source
August 9, 2006
East View of Target
Generated/Stored at Source
SPINDLE-II Overview -- DTN Phase 2 Kickoff
18
© 2006 BBN Technologies
Krishnan
Challenges For ContentBased DTN
Opportunistic Caching (Pull) and Replication (Push)
• Hashing content identifiers to particular caches inadequate for DTN
– stable connectivity to caching infrastructure cannot be assumed
• consistent hashing (Karger et al.) will need significant adaptation for DTN
– need self-organized caching that adapts to changing connectivity
• passive caching of received and transiting content
• active pre-fetching under communication constraints
• Ensuring that relevant content is available when/where it is needed
– under communications and storage constraints
– content can be replicated/cached in full or in network-coded chunks
• Maintaining freshness of cached content
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
19
© 2006 BBN Technologies
Krishnan
Challenges For ContentBased DTN
Distributed Indexing, Search, And Clustering
• Immediate access to search engine farms cannot be assumed
– eventual connectivity to search engine farm is unsatisfactory
• especially if the content itself may be cached locally
• therefore, need distributed querying and retrieval
• Indexing will need to take available caches into account
– continued access to source cannot be assumed
– efficient sharing of indices (content metadata) under disruption
• Security issues for generation and handling of metadata
– data may be encrypted
– source of metadata must be authenticated
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
20
© 2006 BBN Technologies
Krishnan
Challenges For ContentBased DTN
User And Operator Interaction
• Importance, relevance, freshness, and timeliness of content
– description of content to be selectively pushed or pulled
– explicitly expressed as well as learned (caching policies)
• User preferences for handling of cache misses
• Query language and user query interface
– e.g., query-by-example, relational and semantic web query
languages, Boolean combination of keywords
• Appropriate feedback to user (e.g., on mouse-over)
– freshness of content
– expected time to retrieve
– notification when new content becomes available
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
21
© 2006 BBN Technologies
Krishnan
Outline
• BBN’s SPINDLE project
• A declarative knowledge-based approach to DTN
• Late binding of intentional names for endpoints
• Disruption-tolerant access to content
• Other research thrusts: routing and policy support
• Schedule and team
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
22
© 2006 BBN Technologies
Krishnan
Adaptive Routing And Dissemination
Lesson From Phase 1
Offered load: 40 bundles of 2800 bytes; Time:1200s*; Source-Destination path: 6-7 hops
In case of dissemination based strategies, LS updates are sent every 30s
Delivery Ratio
Bundle Forwardings
Zero knowledge strategy
consumes a lot of resources
Hybrid performs
better than pure
strategies
Hybrid performs worse
than zero knowledge
strategy because of
high update rate
Non-adaptive
dissemination does
not scale due to high
update rate
Adaptive dissemination will help us track these curves
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
*Time allowed is not sufficient to deliver all bundles in this 5-x-4 grid DTN scenario
23
© 2006 BBN Technologies
Krishnan
Adaptive Dissemination and Routing
• Adaptive dissemination service shared by routing strategies
– exchange subset of topology information with neighbors
• not quite link state or distance vector
• succinct space-time representation of topology
– control rate, scope, and content of updates
– tradeoff control overhead versus opportunistic data transfer
– service can be used to distribute policy/naming/content metadata too
• Adaptive hybrid routing strategies
– make use of routing information when available
• handle scheduled and predicted connectivity and associated uncertainty
– explore locally when path is unknown or highly uncertain
– choice of strategies based on policy
• multiple routing strategies can be supported via plug-ins
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
24
© 2006 BBN Technologies
Krishnan
Policy Based Resource Utilization
• Declare and control resource use in DTN nodes
– link formation/use based on costs and delivery requirements
• choice of convergence layer parameters
– storage management
– choice of security services
• A declarative language to express and process policies
– node primitives controlled by policy are explicitly declared (ontology)
– deductive database rule execution
• check for policy consistency and conformance of usage to policy
– constraint solver
• search for communication opportunities that are authorized by policy
• Protocol for policy dissemination and consistent use in a DTN
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
25
© 2006 BBN Technologies
Krishnan
Evaluation Platform (From Phase 1)
Approach: Combines OS Virtualization and Emulation
• Multiple real DTN system instances on single machine
• Connect to emulated network via virtual Ethernet bridge
• Flexible scripting of DTN scenarios and traffic, repeatable
Host OS (Linux)
Emulation Script
Set up
nodes
link properties
mobility models
traffic agents
routing
link schedules
dtnd1
dtnd2
dtnd3
User Mode Linux 1
User Mode Linux 2
User Mode Linux 3
Emulation Manager (modified ns-2)
• manage interactions with user-mode linux
 start processes, access interfaces, access dtnd CLI
 ethernet addresses to ns2 node ids
• copy link layer packet to appropriate interface
 after simulated loss/delay and error through network
Analysis
Visualization (nam)
August 9, 2006
Trace
Machine specifications
–
–
–
–
–
–
4 Intel Xeon MP CPU,
2.7GHz, 2MB cache
8GB RAM
300GB SCA SCSI drive
Integrated 10/100 NIC
6 PCI-X slots
16 DIMM Slots (32GB max)
SPINDLE-II Overview -- DTN Phase 2 Kickoff
26
© 2006 BBN Technologies
Krishnan
Outline
• BBN’s SPINDLE project
• A declarative knowledge-based approach to DTN
• Late binding of intentional names for endpoints
• Disruption-tolerant access to content
• Other research thrusts: routing and policy support
• Schedule and team
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
27
© 2006 BBN Technologies
Krishnan
SPINDLE-II Schedule Summary
Period Of Performance: 18 Months Starting July 2006
CORE
WITH ALL
PLUG-IN
INTERFACES
FINAL
DRAFT
PLUG-IN
PLUG-IN
INTERFACE INTERFACE
SPEC
SPEC
BASIC
PLUG-INS
BASELINE CODE FREEZE
FOR
SYSTEM
INTEGRATED PROTOTYPE
MILITARY ADVANCED
CL PLUG-IN PLUG-INS
PROTOTYPES
AVAILABLE
TO PARTNERS
HW REQUIREMENTS
& CANDIDATES
IDENTIFIED
PHASE 2
DEMO
PLAN
RADIO, DATE,
AND VENUE FOR
DEMO IDENTIFIED
PHASE 2
DEMO
INTERNET DRAFTS
(LB, POLICY, OTHER)
DRAFT FINAL
GNG FINAL REPORT
REPORT
0
1
START
Jul 24 2006
2
August 9, 2006
3
4
5
6
7
8
9
10
11
12
13
SPINDLE-II Overview -- DTN Phase 2 Kickoff
14
15
16
17
18
END
Jan 21 2008
28
© 2006 BBN Technologies
Krishnan
SPINDLE-II Team
Dr. Rajesh Krishnan *
(Principal Investigator)
•
•
Dr. Stephen Polit *
(Project Manager)
Dr. Christopher Small *
(System R&D Lead)
Dr. Prithwish Basu *
(Innovations Lead)
Joanne Mikkelson
Christine Jones
John Burgess *
Jeff Ward
Dave Moran
…
Dr. Ram Ramanathan *
Dr. Carl Livadas
Matt Condell
…
Subcontractors and consultants from industry and academia
– in three areas: security, knowledge base, DTN2 internals (being formalized)
Other expertise and support from BBN as needed
– in several areas e.g., transition, security, learning, middleware, contracts
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
* Present at the kickoff meeting today
29
© 2006 BBN Technologies
Krishnan
Summary
• Anticipated BBN contributions:
– open-source modular DTN system software and specifications
• within the DTNRG community process
– system plug-ins
–
–
–
–
• control: routing, naming, and policy management
• convergence layers: selected commercial and DoD network technologies
• persistent storage: basic (FS/DBMS) and advanced (KB)
vehicular DTN system prototype (70 units)
technology for disruption-tolerant access to content
demonstration in a military-relevant scenario
publications and Internet Drafts
• For transition beyond Phase 2, we seek inputs on:
– DoD systems/programs suited for injection of DTN technology
– suitable demonstration venues and applications
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
30
© 2006 BBN Technologies
Krishnan
Discussion
August 9, 2006
SPINDLE-II Overview -- DTN Phase 2 Kickoff
31
© 2006 BBN Technologies