conext student workshop€¦ · Wide range of sensor nets: “plugged in ... Emergency management...
23
1 Networking.... research trends and challenges for the coming decade Jim Kurose Department of Computer Science University of Massachusetts Overview where are we now and how did we get here? challenges on beyond today’s Internet driven by applications driven by fundamentals: network science closing thoughts discussion
Transcript of conext student workshop€¦ · Wide range of sensor nets: “plugged in ... Emergency management...
1
Networking....research trends and challenges for
the coming decade
Jim KuroseDepartment of Computer ScienceUniversity of Massachusetts
Overview
where are we now and how did we get here?challenges on beyond today’s Internet
driven by applicationsdriven by fundamentals: network science
closing thoughtsdiscussion
2
Networking: expanding visions
BISDN OSI
ATMInternet
sensornets
pervasivecomputingp2p,
overlays InternetCDN, pub/sub
2000
1990
Internet
DTNwireless
measurement
Networking: expanding visions
BISDN OSI
ATMInternet
sensornets
pervasivecomputingp2p,
overlays InternetCDN, pub/sub
2000
1990
Internet
wireless
applicationsrule!
measurement DTN
3
Sensor nets: wide range of characteristics
data rates
availablepower
actuation,configurability
power: constrained or “plugged in”?data rate: bit rate, duty cycle?reconfigurability: retasking, retargeting how often?users: single-purpose, many?in-situ or remote?
Wide range of sensor nets: embedded
power constrained sensorsmostly data “push,” re-tasking possiblelow bit rate datanetwork design: from scratch
data rates
availablepower
actuation,configurability
4
Embedded Networked Sensing Apps
embedded micro-sensors, on-board processing, wireless interfaces at very small scale
in-situ sensing: need to “be there,”monitor “up close”
spatially, temporally dense environmental monitoring
seismic structure response
marine microorganisms
contaminant transport
ecosystems, biocomplexity
Slide courtesy of D. Estrin
Wide range of sensor nets: “plugged in”
data rates
availablepower
actuation,configurability
powered radarsrapidly steerable beams: data rates: 2 Mbps - 100 Mbps per radarmultiple data consumersnetwork design space: above IP?
5
Wide range of “sensor nets”: sense-and-response systems
radar/weather
satellite observation(EODIS)
auto traffic monitoring
videosurveillance
vehicle trackingin sensor field
habitat monitoring microclimate
monitoringanimal tracking
network traffic monitoring
Atmospheric sensing: application driver
Today: Sparse, high-power radarsensing gap: Earth curvature effects prevent 72% of the troposphere below 1 km from being observedcoarse resolution
10,000 ft
tornado
wind
earth surface
snow
3.05
km
3.05
km
0 40 80 120 160 200 240RANGE (km)
Horz. Scale: 1” = 50 kmVert. Scale: 1” -=- 2 km
5.4
km
1 km 2
km
4 km
gap
“There is insufficient knowledge about what is actually happening (or is likely to happen) at the Earth’s surface where people live.” [NRC 1998]
6
CASA: collaborative adaptive sensing of the atmosphere
CASA: dense network of low power radars:sense lower 3 km of earth’s atmospherecollaborating radars:
improved sensingimproved detection, prediction
responsive to multiple end-user needs: sample atmosphere when and where end user needs are greatest
10,000 ft
tornado
wind
earth surface
snow
3.05
km
3.05
km
0 40 80 120 160 200 240RANGE (km)
Remote sensingMicrowave engineeringNetworkingReal-time systemsStorageNumerical predictionEmergency managementRadar meteorologyQuantitative inversionClimate studiesSocial impactAntenna design
core partners expertise
A multidisciplinary approac is required!
NSF Engineering Research Center
Sept. 2003
CASACollaborative
Adaptive Sensing
of the Atmosphere
7
Oklahoma 4-Node Test Bed
Cyril
RushSprings
Chickasha
Lawton
OKOneNet(wired)
1 2 3 4 5 6 7 8 9A G3 G3 G3 G3 G3 G3 G3 G3 G3B G3 G3 G3 G3 G3 G3 G3 G3 G3C G3 G3 G3 G3 G3 G3 G3 G3 G3D G3 G3 G3 G3 G3 G3 G3 G3 G3E G3 G3 G3 G3 G3 G3 G3 G3 G3F G3 G3 G3 G3 G3 G3 G3 G3 G3G G3 G3 G3 G3 G3 G3 G3 G3 G3H R1 R1 R2 R2 R1 G3 C2 G3 G3I R1 F1 F2,R1F2,H2 R1 G3 C2 G3 G3J R1 H1,F1 H1,F1 T2,R1 R1 G3 C2 G3 G3K R1 H1 T2,H1T2,R1 R1 G3 G3 G3 G3
query
optimization
First multi radar data, May 9 2006
Puerto Rico, Off-the-Grid Test Bed24 W Ray Marine radar
solar panel
802.11 directionalantenna
VIA EPIA MII6000
no reliance on infrastructuresolar/battery-operated nodesmulti-antenna multi-hop 802.11directional antenna
802.11Multi-hop
8
Research effortsapplication-level congestion controljoint sensing/routing optimization in remote sensing networksarchitecture: separation of control and data background data transfer
TCP-nicedistributed, application routed, data-file-transfer (many-to-1)
optimization: planning, prediction (Kalman filtering) for radar targeting
Architecture overview
streamingstorage
storage
queryinterface
data
End users: NWS,emergencyresponse
Resource planning,optimization
data policy
resource allocation
SNR
Meteorological DetectionAlgorithms
1 2 3 4 5 6 7 8 9A G3 G3 G3 G3 G3 G3 G3 G3 G3B G3 G3 G3 G3 G3 G3 G3 G3 G3C G3 G3 G3 G3 G3 G3 G3 G3 G3D G3 G3 G3 G3 G3 G3 G3 G3 G3E G3 G3 G3 G3 G3 G3 G3 G3 G3F G3 G3 G3 G3 G3 G3 G3 G3 G3G G3 G3 G3 G3 G3 G3 G3 G3 G3H R1 R1 R2 R2 R1 G3 C2 G3 G3I R1 F1 F2, R1 F2,H2 R1 G3 C2 G3 G3J R1 H1,F1 H1,F1 T2,R1 R1 G3 C2 G3 G3K R1 H1 T2,H1 T2,R1 R1 G3 G3 G3 G3
Feature Repository
MC&C: Meteorological command and control
Meteorological Task
Generation
blackboard
1 Mbps (moment)100 Mbps (raw)
30 sec. “heartbeat”
9
Optimized radar targetingwhere to point the radars?: “where user needs are greatest”multiple “users”
different sensing needs (e.g., beam targeting)different utility
policy
environment (SNR) impacts sensing ability
Optimization:sensor targeting and
configuration
policyenv.state
1 2 3 4 5 6 7 8 9A G3 G3 G3 G3 G3 G3 G3 G3 G3B G3 G3 G3 G3 G3 G3 G3 G3 G3C G3 G3 G3 G3 G3 G3 G3 G3 G3D G3 G3 G3 G3 G3 G3 G3 G3 G3E G3 G3 G3 G3 G3 G3 G3 G3 G3F G3 G3 G3 G3 G3 G3 G3 G3 G3G G3 G3 G3 G3 G3 G3 G3 G3 G3H R1 R1 R2 R2 R1 G3 C2 G3 G3I R1 F1 F2, R1 F2,H2 R1 G3 C2 G3 G3J R1 H1,F1 H1,F1 T2,R1 R1 G3 C2 G3 G3K R1 H1 T2,H1 T2,R1 R1 G3 G3 G3 G3
Feature Repository
Meteorological Task
Generation
radar targeting requests
R1 R2
R1 configurations R2 configurations
Optimizing Radar Targeting: incorporating end user considerations
Where to point?
What to optimize?
Find configuration that gives the best quality data for the highest utility tasks (phenomena) at time step k::
( )∑=ttasksCionsconfigurat
CtQktUJ,,
),(,max
Utility – “how important” is task t to the users at time k?
( )∑=ggroups
gg ktUwktU,
,),(
Quality – “how good” is scanning configuration C (distance, coverage,
# radars) for task t?
R1 R2
R1 configurations R2 configurations
defining overall utility in terms of individual group utility
10
Optimizing Radar Targeting: architecture!
What to optimize?
Find configuration that gives the best quality data for the highest utility tasks (phenomena) at time step k::
( )∑=ttasksCionsconfigurat
CtQktUJ,,
),(,maxseparation of “how important,” U(t,k), from how good ,Q(t,C)U(t,k,Q(t,C)) would have been possible but:
complex to solvecomplex to specify and update U(t,k,Q(t,C))“stovepipe” design
Emulator: comparing scan strategiesutility comparison: sit-and-spin versus sectored scan
utility gained according to MC&C-defined utility of scanning under configuration C
( )∑=ttasksCionsconfigurat
CtQktUJ,,
),(,max
time step
utilit
y
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115
MC&C controlled
sit-and-spin
11
“naturally”: group-sensitive utility for each feature (tornado, wind shear, hail core) scanned… and the survey says…..
User feedback:NWS: want “mental movie” scanning “areas of interest” at regular intervalsFeature-based too “jumpy”Need context: scan areas around features
How to define “how important”: Ug(t,k)
User Utility Rules (revised)
interval-based preferences: “do X every Y time units”
1/ 2.5 minYes2+lowesttask sizevelocity over AOI
E3
1 / minYes1lowesttask sizereflectivity over AOI
E21 / minYes1lowest360timeE1
EMs1 / 2.5 minYes1full volumetask sizestormN2
1 / minYes1lowest360timeN1NWS
Samplinginterval
Contig.#radars
ElevationsSectorSelection
Ruletrigger
Rules
12
The really big pictureimportance of user requirements
“It’s the network, stupid”
“It’s the application, stupid”
“It’s the end-user, stupid”
how to embed user requirements into network?sensor networkscontent distributionspecial-purpose overlays
The big picture: again
radar/weather
satellite observation(EODIS)
auto traffic monitoring
videosurveillance
vehicle trackingin sensor field
habitat monitoring microclimate
monitoringanimal tracking
network traffic monitoring
in spite of differences, commonalities as well!
underwatersensing
13
habitat sensing net
atmospheric sensing nets
habitat sensing
atmosp.sensing
geosensing
applications
physical
The big picture: stovepipes or layers
habitat sensing net
habitat sensing
atmosp.sensing
geosensing
applications
physical
atmospheric sensing netsNetwork
Data
The big picture: stovepipes or layers
14
Big picture: supporting new applications –losing the IP hour glass figure?
middle age: a narrowing mind, a widening waist
IP
TCP UDP
Applications
token
radio, copper, fiber
802.11 PPPEth
IP
TCP UDP
Applications
token
radio, copper, fiber
802.11 PPPEth
diffserv
intservmcastmobile
IP“love handles” NAT IPSEC
IP “hourglass” Middle-age IP “hourglass” ?
IP
TCP UDP
Applications
token
radio, copper, fiber
802.11 PPPEth
IP “hourglass”
middle age: a expanding mind, a slim waist
IP
TCP UDP
overlay services
token
radio, copper, fiber
802.11 PPPEth
clientserver
apps
application overlays
Big picture: supporting new applications –losing the IP hour glass figure?
15
Overlays: bringing networking techniques up to the application
Overview
where are we now?challenges: beyond today’s network
driven by applicationsdriven by fundamentals: network science
lessons learned
16
Network theory and practice:… many many challenges
wireless networks: capacity, coveragepricing, economics
utility-based view of protocolsmeasurement:
techniques: sampling, inference, signal analysismanagement
auto-configuration, rapid deployment, resilience to faults, misconfiguration, bugs
overlay networkscompeting levels of underlay/overlay control
mobilitysecurity
Challenge: on beyond the data plane
Fundamental advances here are hard!“efficiency” not always the most important measure
• tradeoff between x-ity, expected performancelittle/no past work on the “X-ities”
adaptabilityreconfigurabilitysecuritymanageability
Q: data plane performance really the major roadblock?“robustness” (non-fragility)complexity of controlmaintainabilityevolvability
the “X-ities”
17
Theme: performance “fragility”
robust optimization: optimize expected performance plus variability measurerouting: choose routes that work well for wide range of traffic matrices
“On optimal routing with multiple traffic matrices,” C. Zhang, Y Liu, W. Gong, J. Kurose, R. Moll, D. Towsley, IEEE Infocom 2005
overlays: ability to optimize routing in overlay compensates for “poor” underlay routing
use protocols, parameter settings that work well for wide range of scenarios
avoid “fragile” solutions for “robust” ones
(Non) fragility: routingaccommodate wide range of conditions, not so highly optimized to fail catastrophically when operating beyond “normal” operating regime.
traditionally: optimize for known TM, but….
uncertainty in TMsTMs change (e.g. time of day)
λ1λ2
λ3
λ4 λ5 λ7
λ8λ9
“robust” routing: performs “well” over range of TMs, even if non-optimally for a given TM
oblivious routing (AT&T)
performance
TM
18
Routing: avoiding fragile solutionsavoiding “fragility” (achieving robustness): routing with
good expected case performanceavoiding “terrible” worst case performance
tradeoff between average and worst case performanceoptimizing weighed sum of expected, worst case performance
AT&T PoP-level topology (400 OD pairs)
AT&T inferred hourly TMs
other changes: link/node failures, impairments, compromise
Robust routing in wireless networks
“robust” with respect to changes in link connectivity (mobility)
similarity with failover/backup routingIdea: use multiple paths at transport layer
Can move flow rate among different transport-layer paths as underlying network paths change
what is right timescale for optimization?joint optimization: each/every topology change?decoupled/layered:
• optimize (network layer) routing at slow time scale• multipath endpoint rate adaption (transport layer) at finer time
scale
19
Layering provides for timescale decomposition
λ1
λ2
λ4
λ3
time
optimizeroutes
optimizerates
optimizerates
optimizerates optimize
routes
Application
Transport
Network
Link
Research Questions:specific routing, multipath transport protocolsperformance penalty for decoupled optimizations?
scenarios (mobility, topology) when penalty is high?function of timescale differencesimilarities with underlay/overlay optimizationinitially investigate via simulation
performance gains for decoupled optimizationsquantify decrease in signaling
20
The “right” level of complexity
time
solutions in use
understanding of problem area
solutions proposed
early middle late
solu
tion
com
plex
ity
Q: What process determines the “right”level of complexity?
[adapted from Hluchyj 2001]
On being the right size“For every type of animal there is a most convenientsize, and a large change in size inevitably carries with it a change of form” [J. Haldane, 1928]
On being the right complexity?For every type of networked system, there is a most convenient complexity of control, and a large change in size or function inevitably carries with it a change of form of control…
21
Overview
where are we now?challenges on beyond today’s Internet
driven by applicationsdriven by fundamentals: network science
closing thoughts
The constancy of change
Q: What will do to today’s (Internet) network what the Internet did to the telephone network?
“Bellheads”: today’s network of the past“IP hourglass heads:” designing tomorrows’ network of the past today?
Q: is a new community needed to think radically?A: not necessarily, but need to think out-of-the-box, driven by application needs
22
BISDN OSI
ATMInternet
sensornets
pervasivecomputingp2p,
overlays InternetCDN, pub/sub
2000
1990
Internet
DTNwireless
measurement
Networking: an exciting time!
Summary: advice to students
lots of successes to be proud of!lots of interesting on-going effortslots of interesting unanswered questions
applications, applications, applicationsfundamental questions with large half life (thinking outside the box)“on beyond data plane”: the “X-ities”disruptive technology push
23
Thanks!
acknowledgements: D. Towsley, M. Ammar, C. Diot, H. Schulzrinne, G. Parulkar,J. Rexford, L. Zhang, several generations of students
slides available: http://gaia.cs.umass.edu/kurose/talks/
