So, there I was, making a presentation to a Scandinavian consulting client regarding the potential benefits of deploying HEVC
or VP9 streams to compatible clients instead of
H.264. It’s an issue I’ve gone round and round
about in my mind because of one data point:
Netflix’s ISP Speed Index, which hovers around
4Mbps for the U.S. and many European countries.
According to Netflix (https://ispspeedindex
. netflix.com), “The Netflix ISP Speed Index is
a measure of prime time Netflix performance
on particular ISPs (internet service providers)
around the globe.” In other words, it’s about
how much data Netflix can push through a particular ISP during prime time.
So, my thinking went, if Netflix can only push
through 4Mbps, then that’s all this particular
client would be able to push through. That being the case, switching from H.264 to VP9/HEVC
wouldn’t reduce bandwidth costs, it would simply substitute a 4Mbps VP9/HEVC stream for
a 4Mbps H.264 stream. The VP9/HEVC stream
would likely have a higher resolution, and might
look better, but there would be no bandwidth
I asked my audience, “What’s your average
distribution throughput during prime time?”
One member of the group said, “about 4Mbps.” I
nodded sagaciously, and concluded, “So, HEVC/
VP9 would improve QoE, but likely wouldn’t reduce bandwidth costs.” There followed a murmur of agreement, and I was about to move on
when another in the group raised his hand.
“I just finished a study of which streams are
viewed by our VOD clients,” he began. “We distribute a mix of SD and HD video, and 84% of
all streams viewed by our clients are the top-quality HD stream, at 8Mbps, and the top-quality SD stream, at 2.5Mbps. The reason our
average throughput is 4Mbps isn’t because
we’re limited by our ISP’s delivery capacity;
it’s because we’re distributing a mix of SD and
HD videos, usually at the top-quality stream for
both. I suspect it’s the same for Netflix, as well.”
Oops. Don’t you hate it when that happens?
Of course, these facts suggested a complete-
ly different conclusion regarding the economic
benefits of deploying HEVC or VP9. Specifically,
not only would it save bandwidth costs—likely
cutting the 8Mbps HD stream down to 4.5Mbps
or so and delivering lesser efficiencies down the
encoding ladder—it would also boost QoE be-
cause lower-bitrate connections would be able to
view higher-resolution, higher-quality streams.
Later, the gent with the statistics was kind
enough to share the stream distribution with
me. While the HD/SD numbers were mixed, of
the clearly HD numbers, 35.93% of the streams
were the top-quality HD file at 8Mbps, while only
2.6% of the streams were the second-tier HD
file at 4.5Mbps. The vast majority of HD streams
were delivered at the highest available rate. The
bottom three video streams, all under 1Mbps,
represented only .45% of all viewed streams.
At least for this producer, the Netflix ISP
Speed Index had no relevance as a predictor of
effective throughput. Interestingly, in its 2017
“State of the Internet” report, Akamai report-
ed an average connection speed of more than
20Mbps in that particular country, which likely
is the more relevant number to consider.
What does this mean? All encoding decisions must relate back to your own distribution
What’s interesting is how many important
decisions this data can drive. For example, my
Scandinavian client was considering reducing
the number of streams in the encoding ladder
since several were retrieved by less than .2% of
all viewers. If bandwidth is highly constricted,
I’d recommend constant bitrate (CBR) or 110%
constrained variable bitrate (VBR) for encoding
to reduce stream variability and the potential
for stream switching forced by data rate spikes
in the stream, rather than changing retrieval
bandwidth. If not, 150%–200% constrained VBR
would deliver slightly better quality with little
risk of deliverability issues. I might even consider capped constant rate factor (CRF) as a
poor man’s per-title encoding schema.
The bottom line is, always make decisions
according to your own log file.
Pushing Streams in Prime Time
Jan Ozer ( firstname.lastname@example.org) is a streaming
media producer and consultant, a frequent contributor to
industry magazines and websites on streaming-related topics,
and the author of Video Encoding by the Numbers. He blogs
frequently at streaminglearningcenter.com.
Comments? Email us at email@example.com, or check
the masthead for other ways to contact us.