Building a Network Performance Data
Building a Network Performance Data
Building a Network Performance Data
Collection Framework with NDT and GIS
Building a Network Performance Data
Collection Framework with NDT and GIS
Seth Peery, Sr. GIS Architect
Virginia Tech Geospatial Information Sciences
April 18, 2011
Virginia Tech Geospatial Information Sciences
Relevance of Performance
Relevance of Performance
Measurement
Measurement
•
Internet benefits (economic development, health,
(
p
,
,
education, etc.) are derived from
effective use of
productive applications
[1], not mere
presence
of
infrastructure
infrastructure.
•
Performance measurement allows us to approximate
the user’s experience with the network and identify
y
what applications are feasible
[1] Gurstein, M. (2003). "Effective use: a community informatics strategy beyond the digital
[ ] , ( ) y gy y g
divide." First Monday 8(12).
Performance Measurement Use Cases
Performance Measurement Use Cases
Target Group Outcomes/Rationale
The publicp • Compare measured to advertised performancep p
• Model and estimate commodity network performance
CAIs • Identify underserved institutions
• Benchmarking and impact assessment for projects like US-UCAN
R&E connectors • Monitor R&E network health
Pl f t i i i
• Plan future provisioning
Supply-side vs. Demand-side
R
t ti
f C
ti it
Supply-side vs. Demand-side
R
t ti
f C
ti it
Representations of Connectivity
Representations of Connectivity
Supply Side
pp y
Demand Side
• Map the core and infer the edges
• Locations of infrastructure
• Map the edges and infer the core
• Spatial distribution of users
• Locations of infrastructure
• Service footprints
• Advertised speeds
• Spatial distribution of users
• Adoption and Digital Divide
• User-initiated performance t t
• Typical speeds
• Focus of SBDD data collection
• Passive/automated testing
tests
• Before SBDD, this was all we had
g
4
Advertised vs. Measured Performance
Advertised vs. Measured Performance
SBDD Max Advertised Download Speed < 200 mbps 200kbps - 768 kbps 768 kbps - 1.5 mbps 1 5 mbps 3 mbps 1.5 mbps - 3 mbps 3 mbps - 6 mbps 6 mbps - 10 mbps 10 mbps - 25 mbps 25 mbps - 50 mbps 50 mbps - 100 mbps 100 mbps - 1 Gbps > 1 Gbps > 1 Gbps
Virginia Tech NDT Test Data Points under 200 kbps 200kbps - 768 kbps 768kbps - 1.5 mbps 1.5mbps - 3 mbps 3mbps - 6 mbps 6mbps - 10 mbps 6mbps - 10 mbps 10mbps - 25 mbps 25mbps - 100 mbps over 100 mbps 0 0.5 1 2 3 4 Miles
Sources of User-Initiated Network Performance Data
Sources of User-Initiated Network Performance Data
Virginia Tech eCorridors Broadband Map
Virginia Tech eCorridors Broadband Map
• Started in 2006 • Combines NDT with a Google Maps front-end • Nearly 5000 data points to date, and growing http://www.ecorridors.vt.edu/maps/broadbandmap
Network Performance Data in GIS
Network Performance Data in GIS
Virginia Data Points under 200 kbps 200kbps - 768 kbps 768kbps - 1.5 mbps 1.5mbps - 3 mbps5 bps 3 bps 3mbps - 6 mbps 6mbps - 10 mbps 10mbps - 25 mbps 25mbps - 100 mbps over 100 mbps 8
Spatial Analysis of Georeferenced
Spatial Analysis of Georeferenced
NDT results
NDT results
P i t b th
l
d
’t t ll
h
•
Points by themselves don’t tell us much
•
Spatial joins to bounded areas can yield
generalized summary information
generalized summary information
•
Getting a sufficiently large sample size is
critical
critical
•
Interpolation holds potential, once certain
constraints are properly modeled
Effect of AccelerateVirginia Campaigns
Effect of AccelerateVirginia Campaigns
Sample Size 0 1 - 5 6 - 10 11 - 25 26 - 100 101 - 644 10
Generalization by Areal Unit
Generalization by Areal Unit
Vi i i D t P i t
Counties Blocks
Virginia Data Points under 200 kbps 200kbps - 768 kbps 768kbps - 1.5 mbps 1.5mbps - 3 mbps 3mbps - 6 mbps 6mbps - 10 mbps 10mbps 25 mbps 10mbps - 25 mbps 25mbps - 100 mbps over 100 mbps Average Download Avg_downlo no data under 200 kbps 200 kbps - 768 kbps 200 kbps - 768 kbps 768kbps - 1.5 mbps 1.5 mbps - 3 mbps 3 mbps - 6 mbps 6 mbps - 10 mbps 10 mbps - 25 mbps 25 mbps - 100 mbps over 100 mbps over 100 mbps
Areal Averages used in Analysis
Areal Averages used in Analysis
g
g
y
y
County-level aggregated map data
County business patterns
12
Interpolation
Interpolation
Kriging Predicted Values 0.002038659 0.002038659 - 0.2 0.2 - 0.768 0.768 - 1.5 1.500000001 - 3 3.000000001 - 6 6.000000001 - 10 10.00000001 - 20 20.00000001 - 92.98016357 92.98016358 - 100
Testing Anchor Institutions: v1
Testing Anchor Institutions: v1
Testing Anchor Institutions: v1
Testing Anchor Institutions: v1
Anchor
Institutions v2 :
Anchor
Institutions v2 :
Institutions v2 :
Internet2 K20
CAIDATA
Institutions v2 :
Internet2 K20
CAIDATA
Project
Project
Research Directions
Research Directions
Research Directions
Research Directions
• Network performance measurements can be aggregated to any desired spatial unit of analysis: county, ZIP code, Census block
• Data can be analyzed over large regions or at the local level
• Objective, user-contributed network performance measurements can be used as inputs to models that quantify many other
can be used as inputs to models that quantify many other outcomes of network connectivity
• Economic Impact
• Education
• Health Care
• Government Performance
• Safety and SecuritySafety and Security
Combining Performance Measurement
Combining Performance Measurement
with Specialized Research Studies
with Specialized Research Studies
eCommerce Survey Respondents
Speed_Download_Mbps 0.000000 - 0.200000 0.200001 - 0.768000 0.768001 - 1.500000 1.500001 - 3.000000 3.000001 - 6.000000 6.000001 - 10.000000 10.000001 - 25.000000 25.000001 - 100.000000 100.000001 - 319.062000 Health IT Adoption 0 12.5 25 50 75 100 Miles E-Commerce usage by Businesses
Seth’s Research Design for Economic Impact
Seth’s Research Design for Economic Impact
g
g
p
p
Challenges of Using Measured
Challenges of Using Measured
Performance Data in Research
Performance Data in Research
•
Removing the measurement infrastructure as
Removing the measurement infrastructure as
a source of error
•
Controlling for temporal and spatial variability
Controlling for temporal and spatial variability
in individual speed tests
•
Understanding the limits to the
g
generalizability of individual tests
Platform Considerations
Platform Considerations
Platform Considerations
Platform Considerations
• Attend Performance WG on
Wednesday at 0730 to learn more
• I2 is a leader in the measurement space due to NDT
• Is “mass market” network performanceIs mass market network performance measurement a fundamentally
different use case?
• The e de facto de acto standard should be opensta da d s ou d be ope
• More fundamentally…
• User-initiated vs. passive / automated testing?
• Moving the test platform closer to the user … regardless of platform
Contact Information
Contact Information
Contact Information
Contact Information
Seth Peery
Seth Peery
Senior GIS Architect, Enterprise GIS
Virginia Tech Geospatial Information Sciences
2060 Torgersen Hall (0197), Blacksburg, VA 24061
(540) 231-2178
sspeery@vt edu
http://gis.vt.edu