The document discusses the adoption and performance comparison of IPv4 and IPv6 in media streaming, highlighting that over 1/4 of requests are made over IPv6. It examines differences in protocol implementation, network speeds, and how media players can leverage both protocols for improved download speeds. The analysis concludes that while IPv4 is generally faster, performance can vary based on time of day and network conditions, indicating the need for further research and development in multi-CDN strategies.

1. IPv4 IPv6 Multi CDN (protocol)
Media Player
Masaaki NABESHIMA
Dec 18, 2019
Copyright (c) kosho.org 1

2. ◼ Over 1 / 4 requests are IPv6
• IPv6 Connection rate
• Akamai Report (Dec 2019) *1
• The United States: 45.8%
• Japan: 31.3%
• IIJ Koshien Report (Aug 2019) *2
• Japan : 24%
Copyright (c) kosho.org 2
Background: IPv6 is REAL
*1:https://www.akamai.com/us/en/resources/our-thinking/state-of-the-internet-report/state-of-the-internet-ipv6-adoption-visualization.jsp
*2: https://eng-blog.iij.ad.jp/archives/4183

3. ◼ Protocol itself
• Almost same for performance
◼ Network Implementation is different
• e.g. Base protocol and Aggregation box is different
• IPv4: PPPoE
• IPv6: IPoE
• Each aggregation box has different capacities
◼ Network Speed is different
• Many people say IPv6 is faster than IPv4
• IPv4 aggregation box is typically full
• IPv6 Aggregation box has some margin
Copyright (c) kosho.org 3
Background: IPv4 and IPv6 are different

4. ◼ Recent Media Player
• The player downloads media chunks by HTTP (HLS, Dash)
• Media chunk: a peace of movie
• The player connects the chunks and generates a full movie
Copyright (c) kosho.org 4
Background: Media Player
Chunk 1 Chunk 2 Chunk 3 Chunk 4 Chunk 5 Chunk 6 Chunk 7 Chunk 8

5. ◼ Basic idea
1, Media Player can use IPv4 and IPv6 same time
• e.g. one at a time
2, Also the Media player can calculate download speeds of each chunks
3, And then, the Media Player can choose the preferable protocol
Copyright (c) kosho.org 5
IPv4 IPv6 Multi CDN (Protocol) Media Player
Media chunks
IPv6 IPv4 IPv6 IPv4 IPv6 IPv4 IPv6 IPv4
9.1 Mbps 5.4 Mbps 9.8 Mbps 6.4 Mbps 9.9 Mbps 4.8 Mbps 5.8 Mbps 8.7 MbpsDownload Speeds
IPv6 IPv6 IPv6 IPv4 IPv6 IPv6 IPv6 IPv4

6. ◼ Player (HTML5)
• Media Playback
• Browser MSE (Media Source Extensions) API
• Statistics
• Browser Web Performance Timing API
• Statistics beacon to server
• REST or MQTT
◼ Analysis System
• REST + static text
• REST + Microsoft Powered BI
• MQTT + Event Broker + Realtime SQL + BI Tool
• Azure: MQTT + IoT Hub + SAQL + Power BI
Copyright (c) kosho.org 6
Sample Implementation

7. http://ipv6.jpcdn.jp/ (only available for ipv6 clients)
• Speed comparison (one at a time mode)
• http://ipv6.jpcdn.jp/4k-comp.html (25Mbps)
• http://ipv6.jpcdn.jp/2k-comp.html (6Mbps)
• http://ipv6.jpcdn.jp/1m-comp.html (1.2Mbps)
• Multi CDN
• http://ipv6.jpcdn.jp/4k-psmcdn.html (25Mbps)
• http://ipv6.jpcdn.jp/2k-psmcdn.html (6Mbps)
• http://ipv6.jpcdn.jp/1m-psmcdn.html (1.2Mbps)
Copyright (c) kosho.org 7
Sample Site

8. ◼ Nov 21 2019 – Dec 16 2019
• Data for analysis
• Sessions: 149
• Clients: 134
• All Data
• Session: 413
• Filter1 (IPv4 chunks > 9 and IPv6 chunks > 9) : 248
• Filter2 (Compress consequent session by same user) : 149
• Use last session
Copyright (c) kosho.org 8
Results

9. ◼ Average Speed
• IPv4 is faster than IPv6 !
IPv4: 7.81MB/s
IPv6: 6.79MB/s
Copyright (c) kosho.org 9
Analysis

10. ◼ QoE Count
Copyright (c) kosho.org 10
Analysis
count
IPv6 faster (> 4) 5 3.36%
IPv6 faster (4 ~ 2) 7 4.70%
IPv6 faster (2 ~ 1.2) 24 16.11%
Almost same 63 42.28%
IPv4 faster (2 ~ 1.2) 31 20.81%
IPv4 faster (4 ~ 2) 11 7.38%
IPv4 faster (> 4) 8 5.37%

11. ◼ Comparison by Hour
Copyright (c) kosho.org 11
Analysis
0.01
0.1
1
10
100
0 4 8 12 16 20 24
IPv6 / IPv4 Speed Rate (log)
Hour Session# IPv6/IPv4
0 7 2.02
1 8 1.04
2 1 1.14
3 1 1.58
4
5
6 2 0.85
7
8 6 1.16
9 2 1.20
10 9 0.84
11 9 0.82
12 9 0.85
13 6 1.03
14 11 1.04
15 15 1.11
16 14 1.25
17 9 0.91
18 2 1.25
19 11 0.80
20 4 1.63
21 4 1.29
22 9 6.30
23 9 1.89
IPv4/IPv6 (log)

12. ◼ Comparison by Hour (2)
Copyright (c) kosho.org 12
Analysis
Hour Session#
Speed Rate
(IPv6/IPv4)
0-3 17 1.48
4-7 2 0.85
8-11 26 0.93
12-15 41 1.04
16-19 36 1.03
20-23 26 3.29

13. ◼ Trend
• IPv4 IPv6 speed are not the same
• Very different
• Time is a factor for speed
• Day time: IPv4 > IPv6
• Night Time: IPv6 > IPv4
◼ Reason ?
• Network Aggregator Congestion
• Home and Office network
• Domestic CDN
• CDN Cache Hit
• Traffic Shaper ??
Copyright (c) kosho.org 13
Discussion

14. ◼ Deep Analysis
• Client network based analysis
• AS (IPv4 IPv6)
• Office, Home
◼ Single server (non CDN) measurement
• CDN uses multiple servers (sometime round robin)
• Not good for this kind of measurement
• Single server (IPv4 IPv6 dual stack)
• Probably we can get better (stable) results
◼ More data
• Enhance exposure of the test site
◼ Big data analysis environment
• We can use IoT backend
Copyright (c) kosho.org 14
Future Work

15. ◼ IPv4 IPv6 speed comparison
• More data: Commercial content
• Need Plug-ins development for major players (JW, video.js, …)
◼ IPv4 IPv6 multi CDN
• Practical use
• Need Plug-ins development for major players (JW, video.js, …)
• Algorithm tuning
• IPv4, IPv6 selection rate (include source detachment)
• Small chunk correction
• …
Copyright (c) kosho.org 15
I need enhancement

16. ◼ IPv4 IPv6 Multi CDN (protocol) Media Player
• Patented in Japan with J-Stream inc, as a player side multi CDN
• Already applied International PCT (Patent Cooperation Treaty) and
foreign country
◼ Partnering
• Research
• IPv4 IPv6 multi Player is practical, but we do not know much of it
• We can make tons of research papers
• Business
• Player is becoming more important than server
• Good player make money
Copyright (c) kosho.org 16
I need Partners !