Motivation Measuring Decentralized Streaming: Background A Case Study of DTube Methodology Analysis

Conclusions Trinh Viet Doan, Tat Pham, Markus Oberprieler, Vaibhav Bajpai

Technical University of Munich (TUM)

IFIP Networking, June 22–25, 2020

1 / 20 Motivation

Background

Methodology

Analysis

Conclusions Motivation

2 / 20 I Most Internet traffic: video streaming, mostly from centralized services

I YouTube 8.7%, Netflix 12.6% of all global downstream traffic as of 2019 [4]

I Proposals of decentralized solutions to counteract centralization

I Decentralized video streaming:

I In the past: P2P video streaming around 2010 I More recently: DTube (2017), PeerTube (2018), LBRY.tv (2020), ...

⇒ Comparing video streaming from centralized and decentralized services using YouTube and DTube

Motivation

Motivation

Background I Increasing concerns about consolidation in the Internet [1, 2, 3] Methodology Analysis

Conclusions

3 / 20 I Proposals of decentralized solutions to counteract centralization

I Decentralized video streaming:

I In the past: P2P video streaming around 2010 I More recently: DTube (2017), PeerTube (2018), LBRY.tv (2020), ...

⇒ Comparing video streaming from centralized and decentralized services using YouTube and DTube

Motivation

Motivation

Background I Increasing concerns about consolidation in the Internet [1, 2, 3] Methodology Analysis I Most Internet traffic: video streaming, mostly from centralized services Conclusions I YouTube 8.7%, Netflix 12.6% of all global downstream traffic as of 2019 [4]

3 / 20 I Decentralized video streaming:

I In the past: P2P video streaming around 2010 I More recently: DTube (2017), PeerTube (2018), LBRY.tv (2020), ...

⇒ Comparing video streaming from centralized and decentralized services using YouTube and DTube

Motivation

Motivation

Background I Increasing concerns about consolidation in the Internet [1, 2, 3] Methodology Analysis I Most Internet traffic: video streaming, mostly from centralized services Conclusions I YouTube 8.7%, Netflix 12.6% of all global downstream traffic as of 2019 [4]

I Proposals of decentralized solutions to counteract centralization

3 / 20 ⇒ Comparing video streaming from centralized and decentralized services using YouTube and DTube

Motivation

Motivation

Background I Increasing concerns about consolidation in the Internet [1, 2, 3] Methodology Analysis I Most Internet traffic: video streaming, mostly from centralized services Conclusions I YouTube 8.7%, Netflix 12.6% of all global downstream traffic as of 2019 [4]

I Proposals of decentralized solutions to counteract centralization

I Decentralized video streaming:

I In the past: P2P video streaming around 2010 I More recently: DTube (2017), PeerTube (2018), LBRY.tv (2020), ...

3 / 20 Motivation

Motivation

Background I Increasing concerns about consolidation in the Internet [1, 2, 3] Methodology Analysis I Most Internet traffic: video streaming, mostly from centralized services Conclusions I YouTube 8.7%, Netflix 12.6% of all global downstream traffic as of 2019 [4]

I Proposals of decentralized solutions to counteract centralization

I Decentralized video streaming:

I In the past: P2P video streaming around 2010 I More recently: DTube (2017), PeerTube (2018), LBRY.tv (2020), ...

⇒ Comparing video streaming from centralized and decentralized services using YouTube and DTube

3 / 20 Motivation

Background

Methodology

Analysis

Conclusions Background

4 / 20 I Leverages variety of decentralized technologies

I Interplanetary File System (IPFS) for video storage

I Decentralized P2P network for storage and delivery of files I IPFS gateway to access IPFS content over HTTP

I Steem blockchain ecosystem for user- and metadata management

I Token rewards to incentivize content contribution

I Parallels to YouTube in terms of user-interaction features, , monetary incentives

Background: DTube

Motivation

Background

Methodology I Decentralized video streaming service Analysis

Conclusions

5 / 20 I Interplanetary File System (IPFS) for video storage

I Decentralized P2P network for storage and delivery of files I IPFS gateway to access IPFS content over HTTP

I Steem blockchain ecosystem for user- and metadata management

I Token rewards to incentivize content contribution

I Parallels to YouTube in terms of user-interaction features, user interface, monetary incentives

Background: DTube

Motivation

Background

Methodology I Decentralized video streaming service Analysis

I Leverages variety of decentralized technologies Conclusions

5 / 20 I Steem blockchain ecosystem for user- and metadata management

I Token rewards to incentivize content contribution

I Parallels to YouTube in terms of user-interaction features, user interface, monetary incentives

Background: DTube

Motivation

Background

Methodology I Decentralized video streaming service Analysis

I Leverages variety of decentralized technologies Conclusions I Interplanetary File System (IPFS) for video storage

I Decentralized P2P network for storage and delivery of files I IPFS gateway to access IPFS content over HTTP

5 / 20 I Parallels to YouTube in terms of user-interaction features, user interface, monetary incentives

Background: DTube

Motivation

Background

Methodology I Decentralized video streaming service Analysis

I Leverages variety of decentralized technologies Conclusions I Interplanetary File System (IPFS) for video storage

I Decentralized P2P network for storage and delivery of files I IPFS gateway to access IPFS content over HTTP

I Steem blockchain ecosystem for user- and metadata management

I Token rewards to incentivize content contribution

5 / 20 Background: DTube

Motivation

Background

Methodology I Decentralized video streaming service Analysis

I Leverages variety of decentralized technologies Conclusions I Interplanetary File System (IPFS) for video storage

I Decentralized P2P network for storage and delivery of files I IPFS gateway to access IPFS content over HTTP

I Steem blockchain ecosystem for user- and metadata management

I Token rewards to incentivize content contribution

I Parallels to YouTube in terms of user-interaction features, user interface, monetary incentives

5 / 20 Background: DTube

Motivation

Background

I Decentralized video streaming service Methodology Analysis I Leverages variety of decentralized technologies Conclusions I Interplanetary File System (IPFS) for video storage

I Decentralized P2P network for storage and delivery of files I IPFS gateway to access IPFS content over HTTP

I Steem blockchain ecosystem for user- and metadata management

I Token rewards to incentivize content contribution

I Parallels to YouTube in terms of user-interaction features, user interface, monetary incentives

Note: Several additions/changes to DTube’s design since beginning of study in early 2019 5 / 20 Motivation

Background

Methodology

Analysis

Conclusions Methodology

6 / 20 I Measure both YouTube and DTube with the same framework/logic

I From user perspective I Possibility to add other services in the future

I Playout of using ExoPlayer

Methodology

Motivation

Background

Methodology

Analysis

Conclusions I Development of (open-source) Android application to measure video streaming

7 / 20 I Playout of videos using ExoPlayer

Methodology

Motivation

Background

Methodology

Analysis

Conclusions I Development of (open-source) Android application to measure video streaming

I Measure both YouTube and DTube with the same framework/logic

I From user perspective I Possibility to add other services in the future

7 / 20 Methodology

Motivation

Background

Methodology

Analysis

Conclusions I Development of (open-source) Android application to measure video streaming

I Measure both YouTube and DTube with the same framework/logic

I From user perspective I Possibility to add other services in the future

I Playout of videos using ExoPlayer

7 / 20 Methodology

Motivation

Background

Methodology

Analysis

Conclusions I Development of (open-source) Android application to measure video streaming

I Measure both YouTube and DTube with the same framework/logic

I From user perspective I Possibility to add other services in the future

I Playout of videos using ExoPlayer

7 / 20 1. Acquire Web page URLs for n videos from “trending” list of each platform 2. Navigate to Web pages, determine source URLs of videos

I Video resolution: 480p

I Different source URL determination based on video service 3. For each video: 3.1 connect() to determined media server (i.e., YouTube media server or DTube IPFS gateway) 3.2 Pass source URL to ExoPlayer for streaming/playout 3.3 Play video for one minute 4. ICMP traceroute measurements to determined media servers 5. Save and upload all measurements, schedule next iteration

Methodology Measurement Process Motivation

Background

Methodology

Analysis

Conclusions

8 / 20 2. Navigate to Web pages, determine source URLs of videos

I Video resolution: 480p

I Different source URL determination based on video service 3. For each video: 3.1 connect() to determined media server (i.e., YouTube media server or DTube IPFS gateway) 3.2 Pass source URL to ExoPlayer for streaming/playout 3.3 Play video for one minute 4. ICMP traceroute measurements to determined media servers 5. Save and upload all measurements, schedule next iteration

Methodology Measurement Process Motivation

Background 1. Acquire Web page URLs for n videos from “trending” list of each platform Methodology Analysis

Conclusions

8 / 20 3. For each video: 3.1 connect() to determined media server (i.e., YouTube media server or DTube IPFS gateway) 3.2 Pass source URL to ExoPlayer for streaming/playout 3.3 Play video for one minute 4. ICMP traceroute measurements to determined media servers 5. Save and upload all measurements, schedule next iteration

Methodology Measurement Process Motivation

Background 1. Acquire Web page URLs for n videos from “trending” list of each platform Methodology Analysis

2. Navigate to Web pages, determine source URLs of videos Conclusions I Video resolution: 480p

I Different source URL determination based on video service

8 / 20 4. ICMP traceroute measurements to determined media servers 5. Save and upload all measurements, schedule next iteration

Methodology Measurement Process Motivation

Background 1. Acquire Web page URLs for n videos from “trending” list of each platform Methodology Analysis

2. Navigate to Web pages, determine source URLs of videos Conclusions I Video resolution: 480p

I Different source URL determination based on video service 3. For each video: 3.1 connect() to determined media server (i.e., YouTube media server or DTube IPFS gateway) 3.2 Pass source URL to ExoPlayer for streaming/playout 3.3 Play video for one minute

8 / 20 Methodology Measurement Process Motivation

Background 1. Acquire Web page URLs for n videos from “trending” list of each platform Methodology Analysis

2. Navigate to Web pages, determine source URLs of videos Conclusions I Video resolution: 480p

I Different source URL determination based on video service 3. For each video: 3.1 connect() to determined media server (i.e., YouTube media server or DTube IPFS gateway) 3.2 Pass source URL to ExoPlayer for streaming/playout 3.3 Play video for one minute 4. ICMP traceroute measurements to determined media servers 5. Save and upload all measurements, schedule next iteration

8 / 20 I Total video duration

I TCP connect time to media server

I Startup delay

I traceroute: IP path length

Methodology Measurement Metrics Motivation

Background

Methodology

Analysis

Conclusions

9 / 20 Methodology Measurement Metrics Motivation

Background

Methodology

Analysis

Conclusions

I Total video duration

I TCP connect time to media server

I Startup delay

I traceroute: IP path length

9 / 20 I February 2019–November 2019 (10 months)

I >8,500 videos measured from both platforms combined

I Over both cellular (LTE) and WiFi (University network)

I Four mobile phones

I Three locations: Munich (DE), Prague (CZ), San Diego (US)

I Four SIM card providers: T-Mobile (DE), Vodafone (DE), o2 (DE), SIMPLE Mobile (US)

Methodology Measurement Experiment Motivation

Background

Methodology

Analysis

Conclusions

10 / 20 Methodology Measurement Experiment Motivation

Background

Methodology

Analysis

I February 2019–November 2019 (10 months) Conclusions

I >8,500 videos measured from both platforms combined

I Over both cellular (LTE) and WiFi (University network)

I Four mobile phones

I Three locations: Munich (DE), Prague (CZ), San Diego (US)

I Four SIM card providers: T-Mobile (DE), Vodafone (DE), o2 (DE), SIMPLE Mobile (US)

10 / 20 I How does decentralized video streaming compare with centralized services in terms of performance?

I How distributed are such decentralized services?

I In which areas can decentralized video streaming be improved?

Motivation

Background

Methodology Analysis Analysis Conclusions

11 / 20 Motivation

Background

Methodology Analysis Analysis Conclusions

I How does decentralized video streaming compare with centralized services in terms of performance?

I How distributed are such decentralized services?

I In which areas can decentralized video streaming be improved?

11 / 20 1.0 DTube 0.8 YouTube 0.6

CDF 0.4 0.2 0.0 0 250 500 750 1000 1250 1500 1750 2000 Content Duration [s]

I YouTube: median 619 sec

I Longform videos (>10 minutes) allow additional advertisements

I DTube: median 323 sec

I Video length and monetization/incentive decoupled

Content Duration

Motivation

Background

Methodology

Analysis

Conclusions

12 / 20 I DTube: median 323 sec

I Video length and monetization/incentive decoupled

Content Duration

1.0 Motivation DTube Background 0.8 YouTube Methodology

0.6 Analysis

CDF 0.4 Conclusions 0.2 0.0 0 250 500 750 1000 1250 1500 1750 2000 Content Duration [s]

I YouTube: median 619 sec

I Longform videos (>10 minutes) allow additional advertisements

12 / 20 Content Duration

1.0 Motivation DTube Background 0.8 YouTube Methodology

0.6 Analysis

CDF 0.4 Conclusions 0.2 0.0 0 250 500 750 1000 1250 1500 1750 2000 Content Duration [s]

I YouTube: median 619 sec

I Longform videos (>10 minutes) allow additional advertisements

I DTube: median 323 sec

I Video length and monetization/incentive decoupled

12 / 20 I YouTube: 22 ms WiFi, 44 ms cellular I DTube: 45 ms WiFi, 107 ms cellular

I 45–60 ms for both platforms and all ISPs, except SIMPLE to DTube (300 ms)

T-Mobile (DE) Vodafone (DE) YouTube o2 (DE) SIMPLE Mobile (US) DTube 1.0 1.0 0.8 0.8 0.6 0.6

CDF 0.4 YouTube (WiFi) CDF 0.4 YouTube (Cellular) 0.2 DTube (WiFi) 0.2 DTube (Cellular) 0.0 0.0 0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 TCP Connect Time [ms] TCP Connect Time [ms]

I 75th percentiles by platform and network type:

I 75th percentiles by cellular provider:

TCP connections to YouTube established in about half the time compared to DTube, although roughly within same order of magnitude (< 100 ms).

TCP Connect Time

Motivation

Background

Methodology

Analysis

Conclusions

13 / 20 I 45–60 ms for both platforms and all ISPs, except SIMPLE to DTube (300 ms)

T-Mobile (DE) Vodafone (DE) YouTube o2 (DE) SIMPLE Mobile (US) DTube 1.0 0.8 0.6

CDF 0.4 0.2 0.0 0 50 100 150 200 250 300 350 TCP Connect Time [ms]

I 75th percentiles by cellular provider:

TCP connections to YouTube established in about half the time compared to DTube, although roughly within same order of magnitude (< 100 ms).

TCP Connect Time

1.0 Motivation 0.8 Background 0.6 Methodology CDF 0.4 YouTube (WiFi) YouTube (Cellular) 0.2 DTube (WiFi) Analysis DTube (Cellular) 0.0 Conclusions 0 50 100 150 200 250 300 350 400 TCP Connect Time [ms]

I 75th percentiles by platform and network type: I YouTube: 22 ms WiFi, 44 ms cellular I DTube: 45 ms WiFi, 107 ms cellular

13 / 20 TCP connections to YouTube established in about half the time compared to DTube, although roughly within same order of magnitude (< 100 ms).

TCP Connect Time T-Mobile (DE) Vodafone (DE) YouTube o2 (DE) SIMPLE Mobile (US) DTube 1.0 1.0 Motivation 0.8 0.8 Background 0.6 0.6 Methodology CDF CDF 0.4 YouTube (WiFi) 0.4 YouTube (Cellular) 0.2 DTube (WiFi) 0.2 Analysis DTube (Cellular) 0.0 0.0 Conclusions 0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 TCP Connect Time [ms] TCP Connect Time [ms]

I 75th percentiles by platform and network type: I YouTube: 22 ms WiFi, 44 ms cellular I DTube: 45 ms WiFi, 107 ms cellular

I 75th percentiles by cellular provider: I 45–60 ms for both platforms and all ISPs, except SIMPLE to DTube (300 ms)

13 / 20 TCP Connect Time T-Mobile (DE) Vodafone (DE) YouTube o2 (DE) SIMPLE Mobile (US) DTube 1.0 1.0 Motivation 0.8 0.8 Background 0.6 0.6 Methodology CDF CDF 0.4 YouTube (WiFi) 0.4 YouTube (Cellular) 0.2 DTube (WiFi) 0.2 Analysis DTube (Cellular) 0.0 0.0 Conclusions 0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 TCP Connect Time [ms] TCP Connect Time [ms]

I 75th percentiles by platform and network type: I YouTube: 22 ms WiFi, 44 ms cellular I DTube: 45 ms WiFi, 107 ms cellular

I 75th percentiles by cellular provider: I 45–60 ms for both platforms and all ISPs, except SIMPLE to DTube (300 ms)

TCP connections to YouTube established in about half the time compared to DTube, although roughly within same order of magnitude (< 100 ms).

13 / 20 I YouTube: 0.82 sec WiFi, 1.35 sec cellular I DTube: 3.2 sec WiFi, 5.8 sec cellular

I YouTube: 1–1.8 sec for all providers I DTube: 3.1–4.6 sec for all providers but SIMPLE (9.8 sec)

T-Mobile (DE) Vodafone (DE) YouTube o2 (DE) SIMPLE Mobile (US) DTube 1.0 1.0 0.8 0.8 0.6 0.6 CDF CDF 0.4 YouTube (WiFi) 0.4 YouTube (Cellular) 0.2 DTube (WiFi) 0.2 DTube (Cellular) 0.0 0.0 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 0 2 4 6 8 10 12 14 Startup Delay [s] Startup Delay [s]

I 75th percentiles by platform and network type:

I 75th percentiles by cellular provider:

Startup delay for DTube about four times higher compared to YouTube; cellular measurements from the US to DTube perform significantly worse.

Startup Delay

Motivation

Background

Methodology

Analysis

Conclusions

14 / 20 I YouTube: 1–1.8 sec for all providers I DTube: 3.1–4.6 sec for all providers but SIMPLE (9.8 sec)

T-Mobile (DE) Vodafone (DE) YouTube o2 (DE) SIMPLE Mobile (US) DTube 1.0 0.8 0.6

CDF 0.4 0.2 0.0 0 2 4 6 8 10 12 14 Startup Delay [s]

I 75th percentiles by cellular provider:

Startup delay for DTube about four times higher compared to YouTube; cellular measurements from the US to DTube perform significantly worse.

Startup Delay

1.0 Motivation 0.8 Background 0.6 Methodology CDF 0.4 YouTube (WiFi) YouTube (Cellular) 0.2 DTube (WiFi) Analysis DTube (Cellular) 0.0 Conclusions 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 Startup Delay [s]

I 75th percentiles by platform and network type: I YouTube: 0.82 sec WiFi, 1.35 sec cellular I DTube: 3.2 sec WiFi, 5.8 sec cellular

14 / 20 Startup delay for DTube about four times higher compared to YouTube; cellular measurements from the US to DTube perform significantly worse.

Startup Delay T-Mobile (DE) Vodafone (DE) YouTube o2 (DE) SIMPLE Mobile (US) DTube 1.0 1.0 Motivation 0.8 0.8 Background 0.6 0.6 Methodology CDF CDF 0.4 YouTube (WiFi) 0.4 YouTube (Cellular) 0.2 DTube (WiFi) 0.2 Analysis DTube (Cellular) 0.0 0.0 Conclusions 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 0 2 4 6 8 10 12 14 Startup Delay [s] Startup Delay [s]

I 75th percentiles by platform and network type: I YouTube: 0.82 sec WiFi, 1.35 sec cellular I DTube: 3.2 sec WiFi, 5.8 sec cellular

I 75th percentiles by cellular provider: I YouTube: 1–1.8 sec for all providers I DTube: 3.1–4.6 sec for all providers but SIMPLE (9.8 sec)

14 / 20 Startup Delay T-Mobile (DE) Vodafone (DE) YouTube o2 (DE) SIMPLE Mobile (US) DTube 1.0 1.0 Motivation 0.8 0.8 Background 0.6 0.6 Methodology CDF CDF 0.4 YouTube (WiFi) 0.4 YouTube (Cellular) 0.2 DTube (WiFi) 0.2 Analysis DTube (Cellular) 0.0 0.0 Conclusions 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 0 2 4 6 8 10 12 14 Startup Delay [s] Startup Delay [s]

I 75th percentiles by platform and network type: I YouTube: 0.82 sec WiFi, 1.35 sec cellular I DTube: 3.2 sec WiFi, 5.8 sec cellular

I 75th percentiles by cellular provider: I YouTube: 1–1.8 sec for all providers I DTube: 3.1–4.6 sec for all providers but SIMPLE (9.8 sec)

Startup delay for DTube about four times higher compared to YouTube; cellular measurements from the US to DTube perform significantly worse. 14 / 20 1.0 0.8 0.6

CDF 0.4 YouTube (WiFi) YouTube (Cellular) 0.2 DTube (WiFi) DTube (Cellular) 0.0 5 7 9 11 13 15 17 19 21 IP Path Length

I By platform and network type: I YouTube: reachable within 10 IP hops (WiFi: 93.9%, cellular: 86.0%)

I DTube: reachable within 10 IP hops only for 4.6% (WiFi); minimum over cellular 11 IP hops (29.8%)

I Paths to YouTube shorter by 7–8 IP hops in comparison

traceroute success rates over cellular network highly depend on ISP. Around 90% of the YouTube destinations within 10 IP hops; for DTube, more than 95% of the destinations beyond 10 IP hops.

I traceroute: failure rate of 17.3% over cellular network

I Highly varying success rate based on cellular ISP

IP Path Lengths

Motivation

Background

Methodology

Analysis

Conclusions

15 / 20 1.0 0.8 0.6

CDF 0.4 YouTube (WiFi) YouTube (Cellular) 0.2 DTube (WiFi) DTube (Cellular) 0.0 5 7 9 11 13 15 17 19 21 IP Path Length

I By platform and network type: I YouTube: reachable within 10 IP hops (WiFi: 93.9%, cellular: 86.0%)

I DTube: reachable within 10 IP hops only for 4.6% (WiFi); minimum over cellular 11 IP hops (29.8%)

I Paths to YouTube shorter by 7–8 IP hops in comparison

traceroute success rates over cellular network highly depend on ISP. Around 90% of the YouTube destinations within 10 IP hops; for DTube, more than 95% of the destinations beyond 10 IP hops.

IP Path Lengths

Motivation

I traceroute: failure rate of 17.3% Background

over cellular network Methodology

I Highly varying success rate based Analysis

on cellular ISP Conclusions

15 / 20 traceroute success rates over cellular network highly depend on ISP. Around 90% of the YouTube destinations within 10 IP hops; for DTube, more than 95% of the destinations beyond 10 IP hops.

IP Path Lengths 1.0 0.8 Motivation

I traceroute: failure rate of 17.3% 0.6 Background

CDF 0.4 YouTube (WiFi) over cellular network YouTube (Cellular) Methodology 0.2 DTube (WiFi) I Highly varying success rate based DTube (Cellular) Analysis on cellular ISP 0.0 5 7 9 11 13 15 17 19 21 Conclusions IP Path Length

I By platform and network type: I YouTube: reachable within 10 IP hops (WiFi: 93.9%, cellular: 86.0%)

I DTube: reachable within 10 IP hops only for 4.6% (WiFi); minimum over cellular 11 IP hops (29.8%)

I Paths to YouTube shorter by 7–8 IP hops in comparison

15 / 20 IP Path Lengths 1.0 0.8 Motivation

I traceroute: failure rate of 17.3% 0.6 Background

CDF 0.4 YouTube (WiFi) over cellular network YouTube (Cellular) Methodology 0.2 DTube (WiFi) I Highly varying success rate based DTube (Cellular) Analysis on cellular ISP 0.0 5 7 9 11 13 15 17 19 21 Conclusions IP Path Length

I By platform and network type: I YouTube: reachable within 10 IP hops (WiFi: 93.9%, cellular: 86.0%)

I DTube: reachable within 10 IP hops only for 4.6% (WiFi); minimum over cellular 11 IP hops (29.8%)

I Paths to YouTube shorter by 7–8 IP hops in comparison

traceroute success rates over cellular network highly depend on ISP. Around 90% of the YouTube destinations within 10 IP hops; for DTube, more than 95% of the destinations beyond 10 IP hops.

15 / 20 YouTube DTube

Google LLC, US MNET-AS, DE O2, CZ T-Mobile, US TDDE-ASN1, DE OVH, FR All All (AS15169) (AS8767) (AS5610) (AS21928) (AS6805) (AS16276) cellular 921 (18.9%) 0 (0%) 0 (0%) 105 (2.2%) 35 (0.7%) 1061 (21.7%) 634 (28.9%) 634 (28.9%) WiFi 3623 (74.2%) 3 (0.1%) 196 (4.0%) 0 (0%) 0 (0%) 3822 (78.3%) 1556 (71.1%) 1556 (71.1%) All 4544 (93.1%) 3 (0.1%) 196 (4.0%) 105 (2.2%) 35 (0.7%) 4883 (100%) 2190 (100%) 2190 (100%)

I YouTube: Videos streamed from nearby ISP ASes (caches) and Google AS

I DTube: All videos streamed from OVH AS16276

All traces to DTube end in OVH AS (FR), while YouTube traces end in ISP caches and Google ASes, indicating locational centralization for DTube.

Destination ASes

Motivation

Background

Methodology

Analysis

Conclusions

16 / 20 DTube

OVH, FR All (AS16276) 634 (28.9%) 634 (28.9%) 1556 (71.1%) 1556 (71.1%) 2190 (100%) 2190 (100%)

I DTube: All videos streamed from OVH AS16276

All traces to DTube end in OVH AS (FR), while YouTube traces end in ISP caches and Google ASes, indicating locational centralization for DTube.

Destination ASes

Motivation YouTube Background

Google LLC, US MNET-AS, DE O2, CZ T-Mobile, US TDDE-ASN1, DE Methodology All (AS15169) (AS8767) (AS5610) (AS21928) (AS6805) Analysis cellular 921 (18.9%) 0 (0%) 0 (0%) 105 (2.2%) 35 (0.7%) 1061 (21.7%) Conclusions WiFi 3623 (74.2%) 3 (0.1%) 196 (4.0%) 0 (0%) 0 (0%) 3822 (78.3%) All 4544 (93.1%) 3 (0.1%) 196 (4.0%) 105 (2.2%) 35 (0.7%) 4883 (100%)

I YouTube: Videos streamed from nearby ISP ASes (caches) and Google AS

16 / 20 All traces to DTube end in OVH AS (FR), while YouTube traces end in ISP caches and Google ASes, indicating locational centralization for DTube.

Destination ASes

Motivation YouTube DTube Background

Google LLC, US MNET-AS, DE O2, CZ T-Mobile, US TDDE-ASN1, DE OVH, FR Methodology All All (AS15169) (AS8767) (AS5610) (AS21928) (AS6805) (AS16276) Analysis cellular 921 (18.9%) 0 (0%) 0 (0%) 105 (2.2%) 35 (0.7%) 1061 (21.7%) 634 (28.9%) 634 (28.9%) Conclusions WiFi 3623 (74.2%) 3 (0.1%) 196 (4.0%) 0 (0%) 0 (0%) 3822 (78.3%) 1556 (71.1%) 1556 (71.1%) All 4544 (93.1%) 3 (0.1%) 196 (4.0%) 105 (2.2%) 35 (0.7%) 4883 (100%) 2190 (100%) 2190 (100%)

I YouTube: Videos streamed from nearby ISP ASes (caches) and Google AS

I DTube: All videos streamed from OVH AS16276

16 / 20 Destination ASes

Motivation YouTube DTube Background

Google LLC, US MNET-AS, DE O2, CZ T-Mobile, US TDDE-ASN1, DE OVH, FR Methodology All All (AS15169) (AS8767) (AS5610) (AS21928) (AS6805) (AS16276) Analysis cellular 921 (18.9%) 0 (0%) 0 (0%) 105 (2.2%) 35 (0.7%) 1061 (21.7%) 634 (28.9%) 634 (28.9%) Conclusions WiFi 3623 (74.2%) 3 (0.1%) 196 (4.0%) 0 (0%) 0 (0%) 3822 (78.3%) 1556 (71.1%) 1556 (71.1%) All 4544 (93.1%) 3 (0.1%) 196 (4.0%) 105 (2.2%) 35 (0.7%) 4883 (100%) 2190 (100%) 2190 (100%)

I YouTube: Videos streamed from nearby ISP ASes (caches) and Google AS

I DTube: All videos streamed from OVH AS16276

All traces to DTube end in OVH AS (FR), while YouTube traces end in ISP caches and Google ASes, indicating locational centralization for DTube.

16 / 20 Motivation

Background

Methodology

Analysis

Conclusions Conclusions

17 / 20 I Limited set of (network-related) measurement metrics

I Limited number of measurement configurations, geographical bias

I DTube as only representative for decentralized video streaming

I Several additions/changes to DTube (video sources, tokens, ...)

Conclusions: Limitations

Motivation

Background

Methodology

Analysis

Conclusions

18 / 20 Conclusions: Limitations

Motivation

Background

Methodology

Analysis

Conclusions I Limited set of (network-related) measurement metrics

I Limited number of measurement configurations, geographical bias

I DTube as only representative for decentralized video streaming

I Several additions/changes to DTube (video sources, tokens, ...)

18 / 20 I Higher connection and startup delays for DTube compared to YouTube

I TCP connect times about twice as high (WiFi 45 ms, cellular 107 ms) I Startup delay about four times higher (WiFi 3.2 sec, cellular 5.8 sec) I IP path lengths higher by 7–8 IP hops

I Locational centralization of DTube

I Private IPFS network, lack of distributed content servers I Low number of videos from public IPFS network/gateways

I However, DTube and decentralized technologies still under development

I Open-source Android app: extensible for other streaming services

App Analysis

[ trinhviet.doan | dat.pham | markus.oberprieler | vaibhav.bajpai ]@tum.de

Conclusions: Summary

Motivation

Background

Methodology

Analysis

Conclusions

19 / 20 I Locational centralization of DTube

I Private IPFS network, lack of distributed content servers I Low number of videos from public IPFS network/gateways

I However, DTube and decentralized technologies still under development

I Open-source Android app: extensible for other streaming services

App Analysis

[ trinhviet.doan | dat.pham | markus.oberprieler | vaibhav.bajpai ]@tum.de

Conclusions: Summary

I Higher connection and startup delays for DTube compared to YouTube Motivation

I TCP connect times about twice as high (WiFi 45 ms, cellular 107 ms) Background I Startup delay about four times higher (WiFi 3.2 sec, cellular 5.8 sec) Methodology I IP path lengths higher by 7–8 IP hops Analysis Conclusions

19 / 20 I However, DTube and decentralized technologies still under development

I Open-source Android app: extensible for other streaming services

App Analysis

[ trinhviet.doan | dat.pham | markus.oberprieler | vaibhav.bajpai ]@tum.de

Conclusions: Summary

I Higher connection and startup delays for DTube compared to YouTube Motivation

I TCP connect times about twice as high (WiFi 45 ms, cellular 107 ms) Background I Startup delay about four times higher (WiFi 3.2 sec, cellular 5.8 sec) Methodology I IP path lengths higher by 7–8 IP hops Analysis Conclusions I Locational centralization of DTube

I Private IPFS network, lack of distributed content servers I Low number of videos from public IPFS network/gateways

19 / 20 I Open-source Android app: extensible for other streaming services

App Analysis

[ trinhviet.doan | dat.pham | markus.oberprieler | vaibhav.bajpai ]@tum.de

Conclusions: Summary

I Higher connection and startup delays for DTube compared to YouTube Motivation

I TCP connect times about twice as high (WiFi 45 ms, cellular 107 ms) Background I Startup delay about four times higher (WiFi 3.2 sec, cellular 5.8 sec) Methodology I IP path lengths higher by 7–8 IP hops Analysis Conclusions I Locational centralization of DTube

I Private IPFS network, lack of distributed content servers I Low number of videos from public IPFS network/gateways

I However, DTube and decentralized technologies still under development

19 / 20 App Analysis

[ trinhviet.doan | dat.pham | markus.oberprieler | vaibhav.bajpai ]@tum.de

Conclusions: Summary

I Higher connection and startup delays for DTube compared to YouTube Motivation

I TCP connect times about twice as high (WiFi 45 ms, cellular 107 ms) Background I Startup delay about four times higher (WiFi 3.2 sec, cellular 5.8 sec) Methodology I IP path lengths higher by 7–8 IP hops Analysis Conclusions I Locational centralization of DTube

I Private IPFS network, lack of distributed content servers I Low number of videos from public IPFS network/gateways

I However, DTube and decentralized technologies still under development

I Open-source Android app: extensible for other streaming services

19 / 20 Conclusions: Summary

I Higher connection and startup delays for DTube compared to YouTube Motivation

I TCP connect times about twice as high (WiFi 45 ms, cellular 107 ms) Background I Startup delay about four times higher (WiFi 3.2 sec, cellular 5.8 sec) Methodology I IP path lengths higher by 7–8 IP hops Analysis Conclusions I Locational centralization of DTube

I Private IPFS network, lack of distributed content servers I Low number of videos from public IPFS network/gateways

I However, DTube and decentralized technologies still under development

I Open-source Android app: extensible for other streaming services

://github.com/tv-doan/ifip-net-2020-app (source code) https://github.com/tv-doan/ifip-net-2020-analysis (artifacts) App Analysis

[ trinhviet.doan | dat.pham | markus.oberprieler | vaibhav.bajpai ]@tum.de 19 / 20 References

Motivation [1] Internet Society, “Internet Society Global Internet Report: Consolidation in the Background Internet Economy,” 2019. Methodology Analysis https://future.internetsociety.org/2019/. Conclusions [2] J. Arkko, B. Trammell, M. Nottingham, C. Huitema, M. Thomson, J. Tantsura, and N. ten Oever, “Considerations on Internet Consolidation and the Internet Architecture,” 2019. https://www.ietf.org/archive/id/ draft-arkko-iab-internet-consolidation-02.txt.

[3] Journal of Cyber Policy, “Special Issue: Consolidation of the Internet,” 2020. https://www.tandfonline.com/toc/rcyb20/5/1.

[4] Sandvine, “Global Internet Phenomena Report,” 2019. https://bit.ly/3cvN5Qi.

20 / 20