A Longitudinal View of YouTube Content Delivery

disparity of achievable benefits over upcoming protocols is essential to
identify bottlenecks in content delivery.

In this talk, we present a longitudinal view of YouTube content delivery,
measured from 100 SamKnows probes connected to dual-stacked networks
representing 66 different origin ASes. Using a 4-year long (since Aug 2014)
dataset, we observe that success rates of streaming a stall-free version of a
video over IPv6 have improved over time. We show that a Happy Eyeballs (HE)
race during initial TCP connection establishment leads to a strong preference
over IPv6. However, even though clients prefer streaming videos over IPv6, we
observe worse performance over IPv6 than over IPv4. The worse performance over
IPv6 is due to the disparity in the availability of Google Global Caches (GGC)
over IPv6. We observe that in cases where GGC nodes are dual-stacked, the path
lengths and latencies are comparable over both address families. However, when
the video is cached by a GGC node over IPv6 only, although paths are almost
always shorter over IPv6, yet, latencies are still lower over IPv4, indicating
room for improvement of content delivery over IPv6.

We witness that most of the Google Global Cache (GGC) node deployments are
reachable within 6 IP hops and within 20 ms from users streaming the videos.
Consequently, GGCs reduce IP path lengths and latencies by up to a third over
IPv4 and by up to a half over IPv6 stressing how content cache deployments
are key to improving content delivery for end-users.

To encourage reproducibility of this work, the youtube test is open-sourced
and the entire dataset is made available to the community.