Video: Doing Server-Side Ad Insertion on Live Sports for 25.3M Concurrent Users

Posted on by Russell Trafford-Jones

Delivering ads successfully is done by some services by having the client insert the different ads, and some by inserting the ads at the server end. The choice of which to use requires knowing your customers and how they are most likely to receive your streams. With the prevalence of ad blockers, businesses find the many customers never see the client-side inserted ads. Inserting ads at the server, therefore allows you to get around this as even the ads look like they are a continuation of the same video feed.

The downside of server-side ad insertion (SSAI), whilst rendering the ads unblockable, restricts the ads you can place. Theoretically, in client-side ad insertion, each user can have their own advert. With SSAI, to do that you would need to create a new stream per user which becomes much more computationally hungry. So the sweet spot comes in between the two where viewers are grouped into categories so that only a few tens of streams, for example, are needed to match ten demographics identified to advertisers. This is known as ‘dynamic SSAI’.

Ashutosh Agrawal took to the stage at the Demuxed SF 2019 conference to explain how Hotstar used dynamic SSAI to deliver targeted ads to their 25 million viewers. As an example of your understanding of your viewers driving your choice of ad-delivery technology, Ashutosh explains that close to 85% of their viewing is on mobile and much of that has marginal reception. In hostile network conditions, the requirement for the player to be downloading ads in the background doesn’t work well since the network can only just about support the live video, so a background download pushes the ABR quality down and could even create pausing and rebuffering. It’s for this reason that Hotstar decided that server-side was the way to go.

Ashutosh takes us through how Hotstar approached this large event. In India, cricket is a very popular game which lasts for up to 8 hours a day. This gives rise to a large number of breaks, over 100, which add up to over an hour’s advertising in total so it’s clear to see why this is a massive opportunity for optimisation. Static ad insertion reacts to SCTE 35 markers inserted. This can work well in the sense that for a 40 seconds SCTE marker, the platform can ad an approx 40-second ad or two 20 second ads. However, it isn’t flexible enough to deal with the times when there are far more people watching than that ad agency has paid for which means that Hotstar would end up delivering more viewers than necessary. It would be better for those viewers to see a different ad, triggered by SCTE 35.

As discussed above, doing SSAI for each person is a scalability and cost nightmare, so we quickly see that Targeted SSAI is the way forward. This allows different cohorts of users to be identified. Each cohort will receive its own virtual feed with their own adverts. We then see the architecture of the system showing how the CDN is used. For scaling, we see that they use a cache rather than a database.

Nginx then gets a namecheck as Ashutosh explains how they provide caching, including an nginx memory cache, to deal with up to 50% of the overall load, shared with the CDN if necessary. He then finishes with a look at the best practices they have learnt and what Ashutosh sees is the future for this technique.

Watch now!
Speaker

Ashutosh Agrawal Ashutosh Agrawal
Evangelist/Architect – CTO’s Office,
Hotstar

Webinar: Multicast ABR opens the door to a new DVB era

Posted on by Russell Trafford-Jones

Now available on demand

With video delivery constituting the majority of traffic, it’s clear there’s a big market for it. ON the internet, this is done with unicast streaming where for each receiver, the stream source has to send another stream. The way this has been implemented using HTTP allows for a very natural system, allied Adaptive Bit Rate (ABR), which means that every when your network capacity is constrained (by the network itself or bandwidth contention), you can still get a picture just at a lower bit rate.

But when extrapolating this system linear television, we find that large audience place massive demands on the originating infrastructure. This load on the infrastructure drives its architects to implement a lot of redundancy making it expensive to run. Within a broadcaster, such loads would be dealt with by multicast traffic but on the internet, Multicast is not enabled. For an IPTV system where each employee had access via a program on their PC and/or a set-top-box on their desk, the video would be sent by multicast meaning that it is the network that was providing the duplication of the streams to each endpoint, not the source.

By combining existing media encoding and packaging formats with the efficiency of point-to-multipoint distribution to the edge of IP-based access networks, it is possible to design a system for linear media distribution that is both efficient and scalable to very large audiences, while remaining technically compatible with the largest possible set of already-deployed end user equipment.

This webinar by Guillaume Bichot which is in place of his talk at the cancelled DVB World 2020 event explains DVB’s approach to doing thus that; combining multicast ordination of content with delivery of an ABR feed, called DVB-mABR.

Video broadcast has been digitised since it’s initial broadcasts in the 30s, and more than once. In Europe, we have seen IP carriage (IPTV) services and most recently the hybrid approach where broadband access is merged into transmitted content with the aim of delivering a unified service to the viewer called HbbTV. Multicast ABR (mABR) defines the carriage of Adaptive Bit Rate video formats and protocols over a broadcast/multicast feed. Guillaume explains the mABR architecture and then looks at the deployment possibilities and what the future might hold.

mABR comprises a multicast server at the video headend. This server/transcaster, receives standard ABR feeds and then encapsulates it into multicast before sending. The decoder does the opposite, removing any multicast headers revealing the ABR underneath. It’s not uncommon for mABR to be combined with HTTP unicast allowing the unicast to pick up the less popular channels but for the main services to benefit from multicast.

Guillaume explores these topics plus whether mABR saves bit rate, how it’s deployed and how it can change in the future to keep up with viewers’ requirements.

Watch now on demand!
Speaker

Guillaume Bichot Guillaume Bichot
Principal Engineer, Head of Exploration
Broadpeak

Video: Let’s be hAV1ng you

Posted on by Russell Trafford-Jones

AV1 is now in use for some YouTube feeds, Netflix also can deliver AV1 to Android devices so we are no longer talking about “if AV1 happens” or “when AV1’s finished”. AV1 is here to stay, but in a landscape of 3 new MPEG codecs, VVC, EVC and LCEVC, the question moves to “when is AV1 the right move?”

In this talk from Derek Buitenhuis, we delve behind the scenes of AV1 to see which AV1 terms can be, more-or-less, mapped to which MPEG terms. AV1 is promoted as a royalty-free codec, although notably a patent pool has appeared to try and claim money from users. Because it’s not reusing ideas from other technologies, the names and specific functions of parts of the spec are both not identical to other codecs, but are similar in function.

Derek starts by outlining some of the terms we need to understand before delving in further such as “Temporal Unit” which of course is called a TU and is analogous to a GOP. Then he moves on to highlight the many ways in which previous DCT-style work has been extended meaning the sizes and types of DCT have been increased, and the prediction modes have changed. All of this is possible but increases computation.

Derek then highlights several major tools which have been added. One is the prediction of the Chroma from the Luma signal. Another is the ‘Constrained Direction Enhancement Filter’ which improves the look of diagonal hard edges. The third is ‘switch frames’ which are similar to IDR frames or, as Derek puts it ‘a fancy P-frame.’ There is also a Multi-Symbolic Arithmetic Codec which is a method of guessing a future binary digit which, based on probability, allows you to encode a subset of the number but just enough to ensure that the algorithm will come out with the full number,

After talking about the Loop Restoration Filter Derek then critiques a BBC article which drew, it seems, incorrect conclusions based on not enabling the appropriate functions needed for good compression and also suggesting that there was not enough information provided for anyone else to replicate the experiment. Derek then finishes with MS-SIM plots of different encoders.

Watch now!
Download the slides.
Speaker

Derek Buitenhuis Derek Buitenhuis
Senior Video Encoding Engineer,
Vimeo

Video: NMOS and ST 2110 Pro AV Roadmap

Posted on by Russell Trafford-Jones

ProAV and Broadcast should be best buddies, but only a relatively few companies sell into both. This is because there are legitimate differences in what we need. That being said, interoperability is a helpful end goal for any industry. Whilst proprietary solutions can help kickstart new technologies and be a positive disruptive force, standardisation is almost always beneficial to the industry in the medium to long term.

Whilst broadcast is happy to live with 4:2:2 colour subsampling in much of its workflow, then deliver in 4:2:0, this is often not an option for ProAV who need to take full 4:4:4 4K at 60fps and throw it on a monitor. Whilst 4:4:4 has, technically been possible over SDI for a while, adoption even in the broadcast market has been small.

There are many opportunities for both industries to learn from each other, but it’s hard to overstate the difference in approach of the SMPTE 2110 and NMOS approach to the problem of media over IP compared to the SDVoE model. The former relies on detailed documentation published publicly for anyone who is willing to buy the standard to implement in any way they see fit be that in software or hardware. The latter specifies a chip which has a documented API that does all of the heavy lifting with no option for self-implementation. The fact that the same chip is used everywhere provides the guarantee of interoperability.

One technology which has bridged the gap between ProAV and broadcast is NDI from Vizrt’s Newtek which uses the same binary software application wherever it’s implemented thus providing, like in SDVoE, the interoperability required. The same is true for SRT although they have just released their first draft for IETF standardisation.

In this talk, PESA CTO Scott Barella examines the many existing standards within ProAV and examines their needs such as HDCP. Whilst HDCP, the High-bandwidth Digital Content Protection mechanism, has often been grappled with by broadcasters, it is at least a standard. And it’s a standard that any vendor will have to deal with if they want their equipment to be widely used in the industry. Similarly the requirement for full-frame rate, full-colour UHD is not simply done within many boxes.

The use of PTP within SMPTE’s ST 2110 suite works perfectly in the studio, is arguably not necessary in much of ‘the cloud’ and is widely considered too complex for a ProAV environment. Scott explains that he has thoughts on how to simplify it to make it more practical and taking into account the different use cases.

Secondary interfaces are crucial in much ProAV whereby USB, RS 232 serial and GPI/GPO need to be transported along with the media. And whilst security and encryption are increasingly important for the broadcast industry as it comes to grips with the fact that all broadcasters are vulnerable to hacking attempts, their requirements are not as stringent as the military’s which drives a notable part of the ProAV market. All of these aspects are being considered as part of the ongoing work the Scott is involved with.

Watch now! and download the presentation
Speaker

Scott Barella Scott Barella
CTO, PESA
AIMS co-chair.