Measuring IPv6 use during Summer Koshien 2019 live stream

By on 20 Dec 2019

Category: Tech matters

Tags: , , ,

Blog home

‘Summer Koshien’ is the National High School Baseball Championship held each year at Koshien Stadium. (Image: Kentaro Iemoto, Flickr)

IIJ (AS2497) is a Japanese ISP that also provides CDN services, including live video streaming. Among the live-streaming events hosted at IIJ, by far the biggest is ‘Summer Koshien‘, the National High School Baseball Championship held at Koshien Stadium. The biannual championships started more than 100 years ago, and have become a symbolic amateur sporting event in Japan.

The live streaming of Summer Koshien is organized by the Asahi Shimbun and Asahi Television Broadcasting Corporation and supported by Sports Bull. During this year’s event, IIJ decided to investigate IPv6 use using the access logs, as inspired by a blog post by Yahoo! JAPAN.

Mobile and PC are the most popular way people stream

This year’s championship was held from 6 to 22 August, with 48 games across 14 game days. The traffic reached 700Gbps during the final game on 22 August, a record high for a single game.

In HTTP live streaming, a video stream is encoded into a sequence of segment files, where newly created segment files are continuously added to the playlist. By watching the playlist, a client player downloads and plays back segment files in sequence.

We used the access logs of the content delivery servers and aggregated the request count and bytes sent from successfully delivered segment files. The unique client count was computed as unique pairs of IP address and user-agent.

We also classified user-agents into ‘mobile’ (smartphones and tablets), ‘PC’ and ‘others’ (game-consoles, TVs, feature phones, bots, unknown). Here, the categories represent client device types, and we cannot identify whether a mobile device is connected to a cellular access network or a Wi-Fi network behind a fixed broadband access network.

Figure 1 shows the daily request count, bytes sent and unique client count, with the breakdown of device types. Because ‘other’ is too small, we will focus on ‘mobile’ and ‘PC’ clients hereafter.

Figure 1 — Daily request count, bytes sent and unique client count.

The number of viewers varies from day to day, influenced by factors such as day of the week, popularity and development of the games. In the figures, holiday and weekends can be identified by the grey background. Because many people watch games on TV at home on weekends, the number of online viewers usually decreases on weekends, as is the case for the three-day holiday weekend from 10 to 12 August. But somehow it increased the next weekend (17-18 August).

The shares of mobile and PC are roughly half-and-half for the request counts and bytes sent, but mobile dominates for the unique client counts, due to the fact that the average viewing time of mobile clients is much shorter than PC clients.

Almost one in four stream via IPv6

Figure 2 shows the ratio of IPv6 in the request count, bytes sent and unique clients, for the total, mobile and PC clients.

The IPv6 ratio of mobile clients is higher than PC and more stable over the days. The IPv6 ratio of PC clients fluctuates from day to day, which makes that of the total clients fluctuate. On average, 24% of the requests, 26% of the bytes sent, and 32% of the unique clients are IPv6.

Figure 2 — IPv6 ratio per client type.

Next, we looked at the top 10 ASes, by matching each client’s IP address to the corresponding origin AS number in the BGP table.

The 3rd column in Table 1 shows the AS’s share in traffic volume against the whole traffic, while the 4th column shows the IPv6 ratio in traffic volume within the AS. For the total clients, the top 4 ASes; — OCN, Softbank, KDDI and DOCOMO — account for 61% of traffic. Moreover, for the mobile clients, the top-4 ASes — Softbank, KDDI, DOCOMO and OCN —account for 77% of traffic.

Table 1 — Top 10 ASes, ordered by traffic volume (bytes sent).

There is a widening gap in the IPv6 ratio among the ‘eyeball ISPs’. Softbank is sticking out with its high IPv6 ratio, especially for mobile clients, where IPv6 reaches almost 60%, suggesting Softbank’s aggressive IPv6 deployment strategy. On the other hand, many eyeball ISPs, especially small ones, do not have IPv6 at all.

(Note: IP-to-AS mapping is somewhat confusing for FLET’S IPv6 IPoE, an option for the IPv6 access methods in NTT’s NGN services. Only eight Virtual Network Enablers (VNEs) aggregate traffic on behalf of other eyeball ISPs, where a source IPv6 address is one from a VNE rather than one from the user’s ISP. However, the impact to Table 1 is limited because major eyeball ISPs are also VNEs. In Table 1, the AS numbers of Softbank, KDDI, BIGLOBE, Asahi-Net and OCN are also used for their VNE services.)

We also investigated the differences between IPv4 and IPv6 in quality of experience such as bit rates and playback delays, but could not find a statistically significant difference. Access logs have limited information on the quality of services and besides, the quality of video playback is influenced more by device models and video-streaming libraries than the network-level quality.

Compared with the report by Yahoo! JAPAN, the average IPv6 ratio in request counts increased from 19% to 24%. However, it is difficult to compare as their mobile device share was only 5%, probably as a result of the auto-playback live-streaming featured on the Yahoo! JAPAN top page, which is popular for PC users.

Events could trigger greater IPv6 deployment

We believe that our data is a snapshot of the IPv6 deployment in transition for Japanese eyeball ISPs. The viewers are certainly skewed for high school baseball games as content which we acknowledge is a segment different from those often reported on by popular content providers such as Google and Facebook. Still, we covered a wide range of users as roughly a million viewers were observed every day.

From our study we have two takeaways: 

  1. IPv6 can be used for content delivery without any special care these days.
  2. There is a widening gap among ISPs and content providers in promoting IPv6, but leading ones have made the change.

From the content side, just enabling IPv6 on edge servers will serve 1/4 of users via IPv6. 

For the ISP side, some ISPs are aggressive in deploying IPv6, looking ahead for simplified operations and facilities, while quite a few ISPs are still hesitant.

As we no longer have technical barriers for IPv6 deployment, a small trigger could make a leap towards IPv6 adoption anytime soon. We are looking forward to the next Summer Koshien!

The original version of this article, written in Japanese, can be found at IIJ Engineers Blog.

Kenjiro Cho is Research Director at Internet Initiative Japan, Inc.

Rate this article

The views expressed by the authors of this blog are their own and do not necessarily reflect the views of APNIC. Please note a Code of Conduct applies to this blog.

Leave a Reply

Your email address will not be published. Required fields are marked *