size scale matters: challenged networks...
TRANSCRIPT
1
Size Scale Matters:Challenged Networks Everywhere?!
Jörg Ott
Technische Universität München
www.cm.in.tum.de
CHANTS Workshop
ACM MobiCom 2017
20 October 2017
2© 2017 Jörg Ott | Chair of Connected Mobility | TUM
CHANTS LandscapeWDTN / CHANTS
AOC
MobiOpp
ExtremeCom
DTN
RG
DTN
WG
Dagstuhl
2017
2015
2010
2005
2000
CC
SDS
SIGCOMM, MobiCom, MobiHoc, MobiSys, …INFOCOM, WoWMoM, PerCom, SECON, …
1998 (IPN)
2003
2013
3© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Challenged Networks haven’t disappeared
(well, maybe some of the esoteric cases are now called ICN J)
they are just becoming more mainstream now
and often appear at different scales
• Flashback
• Scaling dimensions
• Case studies: Classical challenged networks
• When networks become challenged
• Perspectives
4© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Roadmap
5© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Flashback
SystemsApplications
Themes
Topics
6© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Flashback
SystemsApplications
Themes
Topics
• Inter-planetary networking• Military networking• Maritime communications (AUVs, buoys, …)• Vehicular and UAV/drone ad-hoc networks • Sensor networks• Emergency and disaster communication• Networking for developing regions• Social opportunistic networking• Community and DIY networks
7© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Flashback
SystemsApplications
Themes
Topics
• Micro blogging, twitter, e-mail• Web access and video streaming• Chat, blackboard, walkie-talkie, image sharing• Music (and video) sharing• Opportunistic Social Networks:
Safebook, D-Book, fbDTN• Distributing (multimedia) (education) content• Entertainment in crowds• Wildlife, environmental, and geo monitoring
(read: challenged IoT)• Medical monitoring, medical support
8© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Flashback
SystemsApplications
Themes
Topics
• Generic platforms• ION• DTN2• IBR-DTN• SCAMPI + liberouter• k x Haggle• NetInf
• Integrated applications• Twimight• PodNet• Opphos• PirateBox & similar
9© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Flashback
SystemsApplications
ThemesMobility data collection and analytics, mobility modeling – Message forwarding, scheduling, dropping, fragmentation – System and performance modeling – Message Ferrying and data mules –Routing^n – Network and source coding – Content distribution, content caching – MANET and IP-related topics – Efficient neighbor discovery – simulators – Implementation and performance evaluation – Convergence layers – Multi-contact encounters in dense networks – Mutex mechanisms, state synchronization – Security and Privacy –Context awareness – Social networking and communities; modeling, analysis, and exploitation – Data centric networks –Leveraging different radios – integrating cellular and Wi-Fi: Mobile offloading – Energy efficiency and power management –Incentive mechanisms – Resource awareness and management –Congestion control – Crowd computing – Sensor networks and crowd sourcing – Network and device monitoring and management – Time protocols – New flavors of transport or session protocols –(Centralized) control (channels) for content distribution – applications
Topics
• From theory to practice
• Sensor networks and applications
• Low energy systems
• Disaster scenarios and remote areas
• Internet-independent networks and applications
• Content sharing in different ways(point-to-point messaging disappearing)
10© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Some trends
• Flashback
• Scaling dimensions
• Case studies: Classical challenged networks
• When networks become challenged
• Perspectives
11© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Roadmap
(Selected) Scaling dimensions
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 12
Node density
(Path) latency
Message size
Link performanceNetwork reach
Network size
Node / networkdynamics
Connectivity
• Defines: node degree/connectivity and interference• Implications: reachability, routing and resource management
• From energy efficient scanning for scarce contact opportunities• To node selection and congestion avoidance
• Research focus: towards sparse networks• Hybrid and adaptive routing schemes• Special considerations for dense networks
13© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Node density
full meshsparse dense
• Defines: network stability• Implications: contact characteristics, node mixing
• Efficiency requirements for node discovery and data exchange• Routing algorithms and their optimization goals (e.g. spread vs. utility)
• Research focus: scenarios with at least partial motion• Fastest ones are probably the satellite ring road
14© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Node / network dynamics
stationary fast movingmovingpartly moving
15© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Connectivity (Kevin Almeroth, CHANTS 2007)
Continuousconnectivity
• Defines: availability of access to a network backbone• Also: connectivity among a set of nodes
• Example: duty cycling – always-on vs. sleepy sensor nodes
• Implications: instant reach and synchronization• Latency, “wormholes” through the backbone• Forwarding vs. carrying• Nature of the system and protocol design
• Research focus: usually mostly disconnected• From autonomous operation to intermittent (dis)connectivity
Intermittentdisconnectivity
Intermittentconnectivity
Infrequentconnectivity
Autonomousoperation
16© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Link / path latency (+predictability)
ns weeks
• Defines: delivery delay (and control loop latency)• Implications: potential for interactivity and protocol design
• also a function of delivery reliability• from transaction-based protocols to self-contained messaging
• Research focus: tens of seconds to few days• Deterministic as in planetary and satellite schedules• Opportunistic as in pocket-switched networks
hours dayssecs mins
17© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Link performance
bits/min Gb/s
• Considers bit error rate: effective data rate
• Defines: link and path capacity• Implications: delivery efficiency and protocol overhead
• Time needed to get something useful done• Feasibility of redundancy
• Research focus: kbit/s to Mbit/s• From sensor networks and low power nodes to WLAN-based systems• Hybrids for control vs. data channels
Mbit/skbits/s
18© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Message size
Bytes GB
• Defines: feasible payload size• “What can a single message do.”• Part of a sequence vs. self-contained
• Implications: system operation and application design• Protocol design and APIs• Acceptable header overhead and per-message operations• Resource management• Transmission scheduling, multiplexing, and fragmentation
• Research focus: small “packets” to images and videos• From BLE beacons to image sharing to audio/video content delivery
MBKB
19© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Network size
2 1010
• Defines: complexity, potential load, competition• Open vs. closed networks• Limiting to a relevant subset: Geographic area, time window
• Implications: how much a node can know / do about others• History of node contacts and associated information
• e.g. for utility computation• Space-time connectivity graph• Resource management (buffer multiplexing, subgrouping, priorities, …)• Trust and security models
• Research focus: 2 – 103, sometimes 104
103 106
20© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Network reach
Local Global
• Defines: scope of operation• within vs. across social networks• closed vs. open groups
• Implications: feasible routing strategies, latency• flooding vs. opportunistic vs. deterministic routing• relayed or not• Also: feasible trust models
• Research focus: dual• local and neighborhood environments• global and inter-planetary
Inter-planetaryNeighborhood City …
The Internet
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 21
Node density
(Path) latency
Message size
Link performanceNetwork reach
Network size
Node / networkdynamics
Connectivity
• Flashback
• Scaling dimensions
• Case studies: Classical challenged networks
• When networks become challenged
• Perspectives
22© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Roadmap
Case Study 1: Crowd Experience
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 23
Node density
(Path) latency
Message size
Link performanceNetwork reach
Network size
Node / networkdynamics
Connectivity
24© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Opphos (KTH/SICS, 2013)
Sour
ce: h
ttps:
//ww
w.si
cs.s
e/pr
ojec
ts/o
ppho
s
25© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Card-stunt as a Service (KAIST, 2017)
Video and paper available at: https://www.sigmobile.org/mobisys/2017/program.php
Case Study 2: DTN over Aerial Carriers
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 26
Node density
(Path) latency
Message size
Link performanceNetwork reach
Network size
Node / networkdynamics
Connectivity
27© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Leveraging airplanes for bulk data [2009]
• Pick up from / delivery to an airport• Dijkstra routing
28© 2017 Jörg Ott | Chair of Connected Mobility | TUM
System model
29© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Some findings (1): 100 messages
30© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Some findings (2): Flight delays
Case Study 3: Here & Now
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 31
Node density
(Path) latency
Message size
Link performanceNetwork reach
Network size
Node / networkdynamics
Connectivity
• Experience sharingfor the moment
• Ephemeral socialnetwork of peoplenearby
• Content of localrelevance• in space and time
• Example• visiting a theme park
32© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Here & Now
33
Card-based representation in messages
NameContent hashApplication ID
Timestamp
Card idAuthor
Object data
Time-to-liveM
etad
ata
Self-containedobject data
Photo + annotation
Comment
Photo + annotation
Comment
Reply + photo
© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Local anchor: Liberouter
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 34
To avoid fiddling with device-to-device ad-hoc networkingTo provide some notion of stable storageTo bootstrap mobile devices independent of an app store
Multiple interconnected to cover an area
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 35
Helsinki Slush 2014 Milano EXPO 2015
36© 2017 Jörg Ott | Chair of Connected Mobility | TUM
We tried this…
Two meta observations• Synchronization across many devices takes a long time• People don’t like bootstrapping apps to their own devices
• Naïve message forwarding has issues in dense networks• All nodes discover all others• Send messages repeatedly (unless already fully received by a peer)
• First step: reduce connectivity artificially
37© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Part 1: Message synchronization
Full Mesh Limited Random Mesh Star Topology
38© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Measurements (1)• Spreading k messages to 25 nodes• in a single WLAN
• Compose a star topology with the star node as a leader.• Instantiate a new leader and assign part of the clients to it.
• New leader creates a SoftAP / Wi-Fi Direct group.• Keep going recursively until all resources are used.
39
Next step: Divide and Conquer
Star Topology Ch 1 Ch 6
© 2017 Jörg Ott | Chair of Connected Mobility | TUM
40
Next step: Divide and Conquer
ch 1
ch 6 ch 11
ch 36 ch 40 ch 44 ch 48
Leader Followers Wi-Fi network© 2017 Jörg Ott | Chair of Connected Mobility | TUM
41© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Measurements (2)
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320
Frac
tion
Dis
trib
uted
Time [seconds]
Dissemination Process with 1, 2, and 3 Wi-Fi Channels, 50 MB content 3 channels 3 channels (sim) 2 channels 2 channel (sim) 1 channel
• Use the browser (, Luke)
• Extend the captive portal of the liberouter for content access• Web portal page structured according to applications
• Application-independent mechanisms to allow for innovation
• Mobile code to visualize and interact with arbitrary messages• Attached to each message as metadata
• (Use cookies and web storage for store-carry-forward)
42© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Part 2: Enabling legacy devices
43
A sample application object
NameContent hashApplication ID
Timestamp
Context reference…
Object data
Time-to-liveM
etad
ata
Self-containedobject data
Photo + annotation
Comment
Photo + annotation
Comment
Reply + photo
© 2017 Jörg Ott | Chair of Connected Mobility | TUM
44
Application functions: logic + data
App
Internet
f() {…}
Browser
Centralized data storeApplication logic
(backend)
Front-end logic + GUIClient data
NameContent hashApplication ID
Timestamp
Context reference
…
Object data
Time-to-live
Application logic (subset)
© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Localizing cloud functions (1)
ApplicationData Unit
(or Application)
new()
presentation()
reply()
summary()
+
HTML5
Name
Content hash
Application IDMetadata
Object data
Application logic (subset)
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 45
46
Localizing cloud functions (2) d2 d1 c2 b4 b3 b2 a3 c1 a2 b1 a1 Content cache
a1
a2
a3
App a
b1
b2
b3
b4
App b
c2
c1
App c
d2
d1
App d
Grouping content by application identifier
Content summary view (message summary)
Per application view (application presenter)a1 c2 c1
d2
d1
a2 a3b1
b2
b3
b4
Composed web view
© 2017 Jörg Ott
47
Localizing cloud functions (3)
Name
Content hash
Application IDMetadata
Object data
summary()presentation()
reply()new()
Name
Content hash
Application IDMetadata
Object data
summary()presentation()
reply()new()
Name
Content hash
Application IDMetadata
Object data
summary()presentation()
reply()new()
Name
Content hash
Application IDMetadata
Object data
summary()presentation()
reply()new()
Select content
Reply
Summarylayout H
TTP
Presentationlayout H
TTP
Composereply M
sg
Create newobject M
sg
© 2017 Jörg Ott
• Flashback
• Scaling dimensions
• Case studies: Classical challenged networks
• When networks become challenged
• Perspectives
48© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Roadmap
What is making the Internet challenged?
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 49
Node density
(Path) latency
Message size
Link / path perf.Network reach
Network size
Node / networkdynamics
Connectivity
Example: Mobile Access Networks
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 50
Node density
(Path) latency
Message size
Link performanceNetwork reach
Network size
Node / networkdynamics
Connectivity
51© 2017 Jörg Ott | Chair of Connected Mobility | TUM
(Highly) Variable link performanceSource: Netradar.org
Performance = f(location, device, context, demand)= f(location, device, context,
time-of-day, day-of-week, season)
52© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Sample streaming system
Streamingserver
Streamingclient
ControlLoop 1
NetworkCongestion Map
Service ControlLoop 2
Throughput updatesLook-ahead requests
ExpectedThroughput
prediction
53© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Performance prediction and adaptation
t (in sec)!
BW (i
n kb
ps)!
Playback Rate!
Time To Outage (Ttto)!
Channel Capacity!
Receiver Rate!4. Data pre-buffered due to predictive feedback!
Initial !pre-buffering!
Time Duration of Outage (Tgap)!
2.Congestion hole detected! 3. Client schedules
variable transmission rate or switches to lower rate!
1. Look-ahead for holes!
5. Client resets to normal media rate or transmission rate!
Example: Internet for Remote Regions
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 54
Node density
(Path) latency
Message size
Link performanceNetwork reach
Network size
Node / networkdynamics
Connectivity
55© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Remote Internet Access
Internet
Fronthaulnetwork
$$$$
• Low capacity• Less than best effort
(scavenging service)• Implied high latency• Extensive caching• Possibly curated content• Local surrogate services• Time-shifted delivery• Request bundling
Example: IoT
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 56
Node density
(Path) latency
Message size
Link / path perf.Network reach
Network size
Node / networkdynamics
Connectivity
Example: Visible Light Services
© 2017 Jörg Ott | Chair of Connected Mobility | TUM 57
Node density
(Path) latency
Message size
Link performanceNetwork reach
Network size
Node / networkdynamics
Connectivity
• Flashback
• Scaling dimensions
• Case studies: Classical challenged networks
• When networks become challenged
• Perspectives
58© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Roadmap
• Long history with themes to last• Esoteric use cases didn’t succeed
• Many lessons learned to be applied elsewhere• Extreme b/w x delay product• Offline operation• Robustness features
• Current Internet trends:Challenged networks vs. networking challenges?• Cloudification of (mobile) apps and services• Ubiquity of IoT• Expanding the (mobile) network in performance and reach
59© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Some concluding thoughts
SCAMPI
• Teemu Kärkkäinen, Mika Välimaa, Esa Hyytiä, Jörg Ott: Opportunistic Content Dissemination Performance in Dense Network Segments. Proceedings of the ACM MobiCom workshop on Challenged Networks (CHANTS), October 2016.
• Teemu Kärkkäinen, Jörg Ott, Lorenzo Valerio, Paul Houghton, Andrea Passarella: Here & Now: Data-centric Local Social Interactions through Opportunistic Networks. (demo paper) Proceedings of the ACM MobiCom workshop on Challenged Networks (CHANTS), October 2016.
• Teemu Kärkkäinen, Jörg Ott: Liberouter: Towards Autonomous Neighborhood Networking. Proceedings of IEEE/IFIP WONS, March 2014.
• Marcin Nagy, Teemu Kärkkäinen, Jörg Ott: Enhancing Opportunistic Networks with Legacy Nodes. ACM SIGMOBILE Mobile Computing and Communications Review: Volume 18 Issue 3, July 2014.
• Mikko Pitkänen, Teemu Kärkkäinen, Jörg Ott, Marco Conti, Andrea Passarella, Silvia Giordano, Daniele Puccinelli, Franck Legendre, Sacha Trifunovic, Karin Hummel, Martin May, Nidhi Hegde, Thrasyvoulos Spyropoulos: SCAMPI: Service Platform for Social Aware Mobile and Pervasive Computing.ACM SIGCOMM Computer Communication Review, Volume 42 Issue 4, pp 503-508, September 2012.
• Ari Keränen, Jörg Ott: DTN over Aerial Carriers. Proceedings of the 4th ACM MobiCom Workshop on Challenged Networks, Beijing, China, September 2009.
60© 2017 Jörg Ott | Chair of Connected Mobility | TUM
Some references