Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Improving Video Quality by Information Sharing !!Ion Stoica, CTO !(also at UC Berkeley and Databricks) !!
Across many sites
While monitoring more than 4 Billion streams per month
We’ve seen 3 patterns
21.8
6
0
3
6
9
12
15
18
21
24
15.21
9.08
0
2
4
6
8
10
12
14
16
32.39
9.74
0
5
10
15
20
25
30
35
* June 2012
26.8
15.5
0
4
8
12
16
20
24
28
263% more 233% more 67.5% more
* June 2012
73% more
1 Viewers watch more when video is not interruptedUGC TV Channel Sport Aggregator
Buffering Impacted Views (BIV): > 2% buffering or > 5s continuous bufferingnon-BIV BIV
2 Viewers watch more when video is higher definition68
31
0
10
20
30
40
50
60
70
89
46
0
15
30
45
60
75
90
32.64
5.97
0
5
10
15
20
25
30
35
4.2
2.6
0
1
2
3
4
5
* July 2012 * June 2012 Live
46% more 119% more 62% more 447% more
Pre. Channel TV Channel TV Channel News
1 Mbps
Viewers leave and do not return to sites when video fails to start 3
Viewers who experience a single video start-up failure return 54% less
Every step of the delivery chain presents unique challenges to delivering video with low interruptions and high bit-rate
Encoders CDNs ISPs Devices
Inaccurate choice of bit-rates & poor matching of screen sizes & resolution to bit-rates
Wide variation in CDN performance across various ISPs with peak load, type of traffic and time of day
Heterogeneity of screen sizes and rendering performance across PCs, mobile hand-sets, tablets, game consoles, set top boxes, connected TVs …
Great diversity in access link speeds (fiber, cable, dsl, 3G/Wi-fi, corporate networks …)
CDNs Vary in Performance over Geographies and Time
There is no single best CDN across geographies, network, and time
25%
50%
25%
CDN 1
CDN 2
CDN 3
• Metric: buffering ratio• One month aggregated data-set
(2011)§ Multiple Flash (RTMP) customers§ Three major CDNs
• 31,744 DMA-ASN-hour with > 100 streams from each CDN§ DMA: Designated Market Area
• Percentage of DMA-ASN-hour partitions a CDN has lowest buffering ratio
Washington DC (Hagerstown): ASN-CXA-ALL
10% 20% 100% 30% 40% 50% 60% 70% 80% 90%
Washington, DC viewer experience differed greatly… Comcast viewers got the best
streams from CDN 1 51% of the time and only 9% from CDN 2
Washington DC (Hagerstown): VZGNI-TRANSIT (19262)
Verizon users got the best streams from CDN 1 only 17% of the time and 77% from CDN 2
There is no single best CDN in the same geographic region or over time
CDN Streaming Failures Are Common Events" % of stream failures: % of streams with video start failures" Three months dataset (May-July, 2011) for a premium customer
using Flash
CDN (relative) performance varies greatly over time
Performance Changes Minute-by-minute
" For the same ISP, CDN performance varies minute by minute over the course of a day
CDN1 CDN2
Buffering Ra5o
Video Start Failures
Average Bitrate
September 24th, 2013 (1 minute data-‐points over 24 hours)
Different quality metrics no always correlated
Viewer Centric Position
" True end-point sensor to see what the viewer sees and inform the source in real time
Conviva Approach to Optimize Viewer Experience
Real-time Measurement from Every Viewer
Stopped/
Exit
Player
States Joining Playing Buffering Playing
Network/
stream
connection
established
Video
buffer
filled upVideo
buffer
empty
Buffer
replenished
sufficiently
time
User
action
Events
Player
Monitoring
Video download rate,
Available bandwidth,
Dropped frames,
Frame rendering rate, etc.
JoinTime (JT) BufferingRa2o(BR) RateOfBuffering(RB)
RenderingQuality(RQ)
Viewer Centric Position
Continuous Optimization
" True end-point sensor to see what the viewer sees and inform the source in real time
" Adjustments to streams every second to account for local environment and Internet variables
Global View and Policy Control
" Dynamic policy control based on real-time patterns across viewers, affiliates, and networks
Conviva Approach to Optimize Viewer Experience
IP Video Streaming Architecture
Mezzanine Crea5on / Transcoding
CMS / Origin Storage
CDNs: Origin storage HDS fragmenta5on HLS fragmenta5on
Real-‐5me Heartbeats
Real-‐5me Feedback/ac5on
HLS Fragments
HDS/HLS Fragments
Conviva Pla.orm Real-‐5me data ingest and processing Real-‐5me metric computa5on Real-‐5me global op5miza5on Big-‐data analysis infrastructure
SmoothStreaming
CDN 1
CDN 2
Business Policy Input
Pulse: Real-‐5me Visibility for publishers
Content Publisher Team
PMC: Real-‐5me Visibility for CDNs
CDN Team
Offline Analysis and Recommenda5ons
Conviva Team
Mezzanine Crea5on / Transcoding
CMS / Origin Storage
CDNs: Origin storage HDS fragmenta5on HLS fragmenta5on
Real-‐5me Heartbeats
Real-‐5me Feedback/ac5on
HLS Fragments
HDS/HLS Fragments
SmoothStreaming
CDN 1
CDN 2
Business Policy Input
Pulse: Real-‐5me Visibility for publishers
Content Publisher Team
PMC: Real-‐5me Visibility for CDNs
CDN Team
Offline Analysis and Recommenda5ons
Conviva Team
Conviva Pla.orm Real-‐5me data ingest and processing Real-‐5me metric computa5on Real-‐5me global op5miza5on Big-‐data analysis infrastructure
IP Video Streaming Architecture
Conviva Platform
Real-‐5me Alerts
Con5nuous real-‐5me measurements from every client
Real-‐5me and historical Insights
Real-‐5me global op5miza5ons
Inference & Predic5on Engine
Bit Rates
CDNs
Decision Engine
Op5mize viewer performance by selec5ng the best op5on within the set of bit rates and CDNs
Akamai
DMA
ASN
DMA
ASN DMA
ASN
Limelight
Level3 Time of day
Localize issues by region, network, CDN, and 5me
Real-‐5me Global Data Aggrega5on and
Correla5on (Streaming Map-‐reduce)
Historical Data Aggrega5on and Analysis
(Hadoop+Hive+Spark)
Global Inference, Decision & Policy
Engine
Real-‐5me Alerts
Con5nuous real-‐5me measurements from every client
Real-‐5me and historical Insights
Real-‐5me global op5miza5ons
Bit Rates
CDNs Op5mize viewer
performance by selec5ng the best op5on within the set of bit rates and CDNs
Akamai
DMA
ASN
DMA
ASN DMA
ASN
Limelight
Level3 Time of day
Localize issues by region, network, CDN, and 5me
Real-‐5me Global Data Aggrega5on and
Correla5on (Streaming Map-‐reduce)
Historical Data Aggrega5on and Analysis
(Hadoop+Hive+Spark)
Global Inference, Decision & Policy
Engine
Inference & Predic5on Engine Decision Engine
Conviva Platform
Key Idea behind Inference & Prediction Engine
Share quality information across views
Use quality information from existing views to predict§ performance of new viewers at join time§ performance of existing views if we were to switch bitrate or
CDN
Create a model consisting of a set of decision tables
Reactively adapting after failure too late!
Inference and Decision Engines
Inference & Prediction Engine
Spark Cluster
Decision Engine
Update decision tables every 1min (5-8 sec processing)
Make decisions in constant time (
Spark• In-memory computation engine
§ APIs in Scala, Python, Java• Unifies batch, streaming, interactive computations
§ Powerful machine learning (MLlib) and soon graph (GraphX) libraries• Used by tens of companies including Yahoo!, Intel
Spark
Spark
Streaming
Shark SQL
GraphX
MLlib
Streaming
Batch, Interactive
Interactive ,IterativeSophisticated algos.
Use Case 1: Best Starting Bit Rate for Single CDN
Pick the best starting bit rate based on decision tables…• Device• Connection type (3G, 4G,
wifi)• Geo (DMA)• ASN• Protocol• Player version• Etc.
Desktop rate
ASN
rate
ASN
rate
ASN
iPad
iPhone
Quality (e.g., buf ra5o) of desktop users in ASN[1] x rate[1]
Use Case 1: Best Starting Bit Rate for Single CDN
Pick the best starting bit rate based on decision tables…• Device• Connection type (3G, 4G,
wifi)• Geo (DMA)• ASN• Protocol• Player version• Etc.
Desktop rate
ASN
rate
ASN
rate
ASN
iPad
iPhone
For an iPad in ASN[1] select highest bit rate providing good quality (i.e., rate[3])
Use Case 2: Best Starting CDN for Multi-CDNs
Pick the best CDN based on decision tables…• Device• Connection type (3G, 4G,
wifi)• Geo (DMA)• ASN• Protocol• Player version• Etc.
Desktop DMA
ASN
DMA
ASN
DMA
ASN
iPad
iPhone
Desktop DMA
ASN
DMA
ASN
DMA AS
N
iPad
iPhone
CDN2 CDN1
Use Case 2: Best Starting CDN for Multi-CDN
Pick the best CDN based on decision tables…• Device• Connection type (3G, 4G,
wifi)• Geo (DMA)• ASN• Protocol• Player version• Etc.
Desktop DMA
ASN
DMA
ASN
DMA
ASN
iPad
iPhone
Desktop DMA
ASN
DMA
ASN
DMA AS
N
iPad
iPhone
CDN2 CDN1
CDN1 >> CDN2 for ASN[2]xDMA[2]
Assign viewers in ASN[2]xDMA[2]
to CDN1
Use Case 3: CDN Switch on Quality DegradationDesktop DMA
ASN
Desktop DMA
ASN
CDN2 CDN1
0
10
20
30
40
0
1000
2000
3000
Buffer Length Trend
Available Bandwidth Trend
Detect and track buffer drop and bandwidth trends and take ac5on to prevent buffering
Client centric predic=ve algorithms track recent trends in available bandwidth, buffer length, rendering performance … and predict quality problems
Global Inference & Predic=on algorithms track global trends by geography, network (ASN), CDN… iden5fy anomalies and predict quality of views
Use Case 4: Asset Publishing and Caching IssuesCDN1 DMA
Asset
DMA
Asset
DMA
Asset
CDN2
CDN3
Global inference algorithms track individual asset failures by device geography, network (ASN), CDN… and iden5fy any regional anomalies
0 10 20 30 40 50 60
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Asset
% Video
Start Failures
Use Case 5: ISP Saturation
CDN1 DMA
ASN
DMA
ASN
DMA
ASN
CDN2
CDN3
This ASN/DMA is saturated on all three CDNs è Don’t switch CDN. Reduce bit rates and maintain
0
20
40
60
80
100
0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000
Band
width Fluctua5o
n
Peak Concurrent Viewers Network Satura5on Point
Use Case 5: ISP Saturation
CDN1 DMA
ASN
DMA
ASN
DMA
ASN
CDN2
CDN3
This ASN/DMA is saturated on all three CDNs è Don’t switch CDN. Reduce bit rates and maintain
0
20
40
60
80
100
0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000
Band
width Fluctua5o
n
Peak Concurrent Viewers Network Satura5on Point
Use Case 5: ISP Saturation
CDN1 DMA
ASN
DMA
ASN
DMA
ASN
CDN2
CDN3
This ASN/DMA is saturated on all three CDNs è Don’t switch CDN. Reduce bit rates and maintain
0
10
20
30
40
0 5000 10000 15000 20000 25000 30000 35000
0
20
40
60
80
100
0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000
Band
width Fluctua5o
n Ba
ndwidth Fluctua5o
n
Peak Concurrent Viewers
Peak Concurrent Viewers
Network Satura5on Point
Unsaturated Network
Challenges" What happens if a partition doesn’t has enough data?
1) Spatial aggregation
§ Which partition use for prediction?
?
CDN1 (buffering raBo)
DMA
ASN
Challenges" What happens if a partition doesn’t has enough data?
1) Spatial aggregation
§ Which partition use for prediction?
2) Temporal aggregation (increase window)
§ What window size?
?
CDN1 (buffering raBo)
DMA
ASN
CDN1 (buffering raBo)
DMA
ASN
Example 1: CDN and Bit Rate Selection
Example 1: Impact on CDN Usage" ASN 7922 (Comcast) – for all of Aug 19th
Metric Buffering Ra=o Avg. Bitrate Video failures
CDN1 1.26 % 2270 K 9.3%
CDN2 1.08 % 2668 K 8.12%
CDN3 1.47 % 2451 9.9%
Metric Buffering Ra=o Avg. Bitrate Video failures
CDN1 2.12 % 1832 K 10.7%
CDN2 2.46 % 1874 K 11.0%
CDN3 2.09 % 1830 K 9.6%
" ASN 20057 (AT&T Wireless) for all of Aug 19th
Precision picks CDN2 64% of the 5me
Precision picks CDN1 & CDN3 74% of the 5me
Example 1: Aggregate over 5 Days
Metric Non-‐Precision Precision Precision Improvement
Buffering Ra5o 2.3 % 1.5 % 33%
Average Bitrate 1692 K 2287 K 35%
Failures and Exits 11.5% 10.6 % 8% Buffering Impacted Views
13.4% 9.4 % 30%
Example 2: Preserving Quality in Presence of Failures!" Precision ensures that audience quality is NOT impacted by CDN failures" Content brands and audience are protected " Content owners can be more aggressive in using capacity from CDN vendors
With Precision Video, CDN problem has no effect on viewers
Without Precision Video, CDN problem has big effect on viewers
Summary" Key transition of main-stream video to the Internet
" Video quality presents opportunity and challenge" Premium video on big screens à zero tolerance for poor quality
" Ability to infer and predict viewer quality key to maximize
quality perceived by users" Reacting after the fact too late!
Conviva Precision Protects Brands, Audience & CDN!" Precision ensures that audience quality is NOT impacted by CDN failures" Content brands and audience are protected " Content owners can be more aggressive in using capacity from CDN vendors
With Precision Video, CDN problem has no effect on viewers
Without Precision Video, CDN problem has big effect on viewers
Other VoD Entertainment Competitors
Unint
erru
pted
View
s (%
)
Average Bitrate (kbps)
Conviva Precision: Simultaneously Increasing Resolution & Reducing Buffering
60
65
70
75
80
85
90
95
100
500 1000 1500 2000 2500
Highest Quality/Most Engaged Audience
(before Precision)
(aker Precision)
With Conviva Precision, Viewers Watch More, Come Back More Often
Before AfterConviva Conviva
Event Info EventDates Feb Mar
Views 41,652 49,607Uniques 9,930 11,448Viewed Minutes 855,002 1,160,667Minutes per View 20.5 23.4Minutes per Unique 86.1 101.4
Quality
Audience
Improvement
19%15%36%14%18%
Increased average bitrate !from 1.7 mbps to 2.1 mbps
Reduced views !impacted by buffering !from 16.13% to 5.56%
Raised engagement by 36%
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
1600170018001900200021002200
Views Impacted by Buffering
Average Bit Rate
Before AfterConviva Conviva
Event Info EventDates Feb Mar
Views 41,652 49,607Uniques 9,930 11,448Viewed Minutes 855,002 1,160,667Minutes per View 20.5 23.4Minutes per Unique 86.1 101.4
Quality
Audience
Improvement
19%15%36%14%18%
First Full Month Results
Lift with Precision
The Truth" Video delivery over the internet is hard
§ CDN variability makes it nearly impossible to deliver high quality all the time with just one CDN
SIGCOMM ‘12
The Idea" Where there is heterogeneity, there is room for
optimization" For each viewer we want to decide what CDN to stream
from" But it’s difficult to model the internet, and things can
rapidly change over time" So we will make this decision based on the real-time data
that we collect
The Truth" Video delivery over the internet is hard
§ CDN variability makes it nearly impossible to deliver high quality everywhere with just one CDN
SIGCOMM ‘12
The Truth" Video delivery over the internet is hard
§ CDN variability makes it nearly impossible to deliver high quality all the time with just one CDN
SIGCOMM ‘12
The Idea" Where there is heterogeneity, there is room for
optimization" For each viewer we want to decide what CDN to stream
from" But it’s difficult to model the internet, and things can
rapidly change over time" So we will make this decision based on the real-time data
that we collect
CDN1 (buffering raBo)
CDN2 (buffering raBo)
The Idea
Avg. buff ra5o of users in Region[1] streaming from
CDN1
Avg. buff ra5o of users in Region[1] streaming from
CDN2
" For each CDN, partition clients by City
" For each partition compute Buffering Ratio
Seal
le San
Francisco
New
York
Lond
on Hon
g Ko
ng
The Idea" For each partition select best CDN
and send clients to this CDN
CDN1 (buffering raBo)
CDN2 (buffering raBo)
Seal
le San
Francisco
New
York
Lond
on Hon
g Ko
ng
The Idea" For each partition select best CDN
and send clients to this CDN
CDN1 (buffering raBo)
CDN2 (buffering raBo)
CDN2 >> CDN1
Seal
le San
Francisco
New
York
Lond
on Hon
g Ko
ng
The Idea" For each partition select best CDN
and send clients to this CDN
CDN1 (buffering raBo)
CDN2 (buffering raBo)
Seal
le San
Francisco
New
York
Lond
on Hon
g Ko
ng
Best CDN (buffering raBo)
The Idea
CDN1 (buffering raBo)
CDN2 (buffering raBo)
" For each partition select best CDN and send clients to this CDN
Seal
le San
Francisco
New
York
Lond
on Hon
g Ko
ng
The Idea
Best CDN (buffering raBo)
CDN1 (buffering raBo)
CDN2 (buffering raBo)
" What if there are changes in performance? S
eall
e San
Francisco
New
York
Lond
on Hon
g Ko
ng
The Idea
Best CDN (buffering raBo)
CDN1 (buffering raBo)
CDN2 (buffering raBo)
" Use online algorithm respond to changes in the network. S
eall
e San
Francisco
New
York
Lond
on Hon
g Ko
ng
Content owners (CMS & Origin)
CDN 1
CDN 2
CDN 3
Clients
Coordinator
Con5
nuou
s measuremen
ts
Busin
ess P
olicies
control
How?" Coordinator implementing an optimization algorithm that
dynamically selects a CDN for each client based on§ Individual client§ Aggregate statistics§ Content owner policies
" All based on real-time data
What processing framework do we use?" Twitter Storm
§ Fault tolerance model affects data accuracy§ Non-deterministic streaming model
" Roll our own§ Too complex§ No need to reinvent the wheel
" Spark§ Easily integrates with existing Hadoop architecture§ Flexible, simple data model§ Writing map() is generally easier than writing update()
Spark Usage for Production
Decision Maker Decision Maker Decision Maker
" Compute performance metrics in spark cluster
" Relay performance information to decision makers
Spark Cluster Spark Node
Spark Node
Spark Node
Spark Node Spark
Node
Storage Layer
Clients
Results
75%
80%
85%
90%
95%
100%
0 500 1000 1500 2000 2500
Non-‐buffering views
Average Bitrate (Kbps)
Spark’s Role" Spark development was incredibly rapid, aided both by its excellent
programming interface and highly active community" Expressive:
§ Develop complex on-line ML decision based algorithm in ~1000 lines of code
§ Easy to prototype various algorithms" It has made scalability a far more manageable problem" After initial teething problems, we have been running Spark in a
production environment reliably for several months.
Problems we faced" Silent crashes…" Often difficult to debug, requiring some tribal knowledge" Difficult configuration parameters, with sometimes inexplicable
results" Fundamental understanding of underlying data model was essential
to writing effective, stable spark programs
Enforcing constraints on optimization" Imagine swapping clients until an
optimal solution is reached
Constrained Best CDN (buffering raBo)
CDN1 (buffering raBo)
CDN2 (buffering raBo)
50%
50%
Enforcing constraints on top of optimization" Solution is found after clients have already joined." Therefore we need to parameterize solution to clients
already seen for online use." Need to compute an LP on real time data" Spark Supported it
§ 20 LPs§ Each with 4000 decisions variables and 350 constraints§ 5 seconds.
Tuning
" Can’t select a CDN based solely on one metric.§ Select utility functions that best predict engagement
" Confidence in a decision, or evaluation will depend on how much data we have collected§ Need to tune time window§ Select different attributes for separation
Tuning
" Need to validate algorithm changes quickly" Simulation of algorithm offline, is essential
Spark Usage for Simulation
Spark Node
Spark Node
Spark Node
Spark Node
Spark Node
HDFS with Produc5on traces
" Load production traces with randomized initial decisions
" Generate decision table (with artificial delay)
" Produce simulated decision set" Evaluate decisions against
actual traces to estimate expected quality improvement
Future of Spark and Conviva" Leverage spark streaming" Unify live and historical processing" Develop platform to build various processing ‘apps’ (e.g.
Anomaly Detection, Customer Tailored Reporting)§ Can share the same data API§ Will all have consistent input
In Summary" Spark was able to support our initial requirement of fast
fault tolerant performance computation for an on-line decision maker
" New complexities like LP calculation ‘just worked’ in the existing architecture
" Spark has become an essential tool in our software stack
Real-time Measurement from Every Viewer
Stopped/
Exit
Player
States Joining Playing Buffering Playing
Network/
stream
connection
established
Video
buffer
filled upVideo
buffer
empty
Buffer
replenished
sufficiently
time
User
action
Events
Player
Monitoring
Video download rate,
Available bandwidth,
Dropped frames,
Frame rendering rate, etc.
JoinTime (JT) BufferingRa2o(BR) RateOfBuffering(RB)
RenderingQuality(RQ)
An UK ISP: Top Metros by Metric over timeGreater London
Op5mizing for Buffering Ra5o
Op5mizing for Video Start Failures
Slough
Op5mizing for Average Bitrate
Birmingham
Greater London
Greater London
Slough
Slough
Birmingham
Birmingham
Time (1 min data-‐points over 24 hours)
Measurements of 3 Leading CDN Show Significant Variation Over Geo and ISP
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
100.0%
Washington DC
(Hagerstow
n):CMCS(33657)
Milw
aukee:RO
ADRU
NNER
-‐CEN
TRAL(20231)
Green Ba
y -‐ A
ppleton:SCRR
-‐7015(7017)
Denver:ASN
-‐QWEST(20
9)
Charlole:SCRR
-‐11426(11426)
Washington DC
(Hagerstow
n):ASN
-‐CXA
-‐ALL-‐
Philade
lphia:VZ
GNI-‐T
RANSIT(19262)
San Diego:SBIS-‐AS(7132)
Las V
egas:ASN
-‐CXA
-‐LV-‐13432-‐CB
S(13432)
Madiso
n:CH
ARTER-‐NET-‐HKY-‐NC(20115)
Indianapolis:SBIS-‐AS(7132)
Providen
ce -‐ New
Bed
ford:ASN
-‐CXA
-‐ALL-‐CCI-‐22773-‐
Washington DC
(Hagerstow
n):VZG
NI-‐T
RANSIT(19262)
Harrord & New
Haven
:SBIS-‐AS
(7132)
Houston:SBIS-‐AS(7132)
Grand Ra
pids -‐ Kalamazoo
-‐ Ba
lle Creek:SBIS-‐
Atlanta:BE
LLSO
UTH
-‐NET-‐BLK(6389)
Hono
lulu:HAW
AIIAN-‐TELCO
M(36149)
Atlanta:CO
MCA
ST-‐7725(7725)
Washington DC
(Hagerstow
n):SPC
S(10507)
Orla
ndo -‐ D
aytona Beach -‐ Melbo
urne
:EMBA
RQ-‐
Denver:CMCS(33652)
Norfolk -‐ Po
rtsm
outh -‐ New
port New
s:AS
N-‐CXA
-‐ALL-‐
Saint Lou
is:CH
ARTER-‐NET-‐HKY-‐NC(20115)
Cincinna5:FU
SE-‐NET(6181)
Phoe
nix:AS
N-‐QWEST(209)
Dallas -‐ Fort W
orth:VZG
NI-‐T
RANSIT(19262)
Pilsburgh:CCC
H-‐3(7016)
Saint Lou
is:SBIS-‐AS(7132)
San Francisco -‐ O
akland
-‐ San Jose:SBIS-‐AS
(7132)
San Diego:RO
ADRU
NNER
-‐WEST(20001)
Greenville -‐ Spartansburg -‐ A
sheville -‐
Kansas City
:SBIS-‐AS
(7132)
Columbu
s -‐ OH:SCRR
-‐10796(10796)
Louisville:INSIGH
T-‐CO
MMUNICAT
IONS-‐CO
RP-‐
Chicago:AT
T-‐INTERN
ET3(6478)
Kansas City
:ASN
-‐CXA
-‐ALL-‐CCI-‐22773-‐RDC
(22773)
Miami -‐ Fort Laude
rdale:BE
LLSO
UTH
-‐NET-‐BLK(6389)
Cleveland:NEO
-‐RR-‐CO
M(11060)
Chicago:VZ
GNI-‐T
RANSIT(19262)
Columbia -‐ SC:SCRR
-‐11426(11426)
Dallas -‐ Fort W
orth:SBIS-‐AS
(7132)
Detroit:S
BIS-‐AS
(7132)
Kansas City
:SCR
R-‐11955(11955)
Los A
ngeles:CHA
RTER
-‐NET-‐HKY-‐NC(20115)
Miami -‐ Fort Laude
rdale:CO
MCA
ST-‐20214(20214)
Cleveland:SCRR
-‐10796(10796)
West P
alm Beach -‐ Fort
San Diego:AS
N-‐CXA
-‐ALL-‐CCI-‐22773-‐RDC
(22773)
Seal
le -‐ Tacoma:AS
N-‐QWEST(209)
New
York:CA
BLE-‐NET-‐1(6128)
Portland
-‐ OR:AS
N-‐QWEST(209)
Oklahom
a City:SBIS-‐AS
(7132)
Phoe
nix:AS
N-‐CXA
-‐ALL-‐CCI-‐22773-‐RDC
(22773)
Chicago:SBIS-‐AS(7132)
Greensbo
ro -‐ High Point -‐ Winston
-‐Albany -‐ Sche
nectady -‐ T
roy:RR
-‐NYSRE
GION-‐
Tyler -‐ Lon
gview (Lut
in &
Grand Ra
pids -‐ Kalamazoo
-‐ Ba
lle Creek:CMCS(33668)
Houston:CM
CS(33662)
Harrord & New
Haven
:COMCA
ST-‐7015(7015)
Eugene
:ASN
-‐QWEST(209)
San Francisco -‐ O
akland
-‐ San Jose:CMCS(33651)