JW Player's Dave LaPalomento Talks VP9, H.264 and More
JW Player is a global product and service provider that enables customers like Fox, Viacom, Univision, and Eurosport to play, stream, monetize and engage their viewers. With a global footprint of over 1 billion unique viewers, JW Player streams a lot of videos to a lot of viewers. JW Player is also a pioneer in video delivery, one of the first online video platforms to deploy the VP9 codec and one of the first to use per-title technology.
That's why we were eager to hear the views of CTO David LaPalomento on codec usage, the future of per-title technology, dynamic and static packaging, and the challenges that advertising insertion places on streaming delivery.
Streaming Media: I'm here today with David LaPalomento, CTO of online video platform JW Player. How are you today, David?
LaPalomento: I'm doing great. How are you doing, Streaming Media?
Streaming Media: I'm good. Take a moment and tell us about JW Player, your positioning and clients that you serve.
LaPalomento: Yeah, absolutely. So JW Player is a complete video platform for publishers, for broadcasters and media companies. And I think we have started years and years back as the earliest way to get video playing on the Web. And really had a incredible following from developers and other media companies around JW Player.
We've expanded into full platforms that includes transcoding and delivery and storage, live video and all sorts of tools and solutions for people who want to deliver video over the Internet.
Streaming Media: Tell us about your training. How did you were you start and how did you get into compression?
LaPalomento: Yeah, well, so it's funny. I kind of I guess I kind of stumbled into video. I was a developer looking for a challenging field where you could be set with hard problems and got involved with an effort years and years ago to turn Flash video players into HTML. I kind of fell in love with the engineering problems that we have to face every day in the video realm, and particularly digital video.
So, I spent a bunch of years doing that when Media Source Extensions came around. I was one of the early people playing around with using that to actually deliver HLS video on the web, which is now a pretty foundational component of most people's web presence, at least.
I have fallen in love with all of the technical challenges we have. We are in a space which is uniquely challenging in terms of technology and still keeps me excited.
Streaming Media: What's your degree in?
LaPalomento: Computer Science.
Streaming Media: How much of what you learned then is relevant to what you do now?
LaPalomento: That's a good question. Let's see how much of it is relevant. So, I think there's still aspects of like analyzing algorithms and systems that is useful. It's nice to have the vocabulary to talk about it, but I don't and can't beat regular expressions. I think that's something I was interested to do is do in university.
It's still an important part of my job. But yeah, I think it's good building blocks. Do I use the building blocks every day? Probably not. Probably not. There's a lot of reality of engineering that goes on day to day, which, you know, is higher layers above that stuff.
Streaming Media: So, JW is notable for many reasons. But from a compression standpoint, you were one of the first to deploy VP9 and one of the first to deploy capped CRF. What codecs are you using today for live and VOD?
LaPalomento: So, I would say H.264 leads the pack still. We have VP9 available to customers and we do have a bunch of customers who have deployed it. But it's more complex to deploy VP9 still today, and there are tradeoffs that you have to consider if you wanted to play it.
So really, it's our sophisticated customers who tend to take advantage of VP9 in their actual deployments. and most people get by and do very well with H.264 still.
Streaming Media: So, what are the tradeoffs as VP9?
LaPalomento: Well, a big tradeoff is that it takes longer and it's more computationally expensive. The other factor in is that you are storing extra renditions or of your content. And then so that adds to your storage bill, even if it's a smaller actual file size.
The other thing to think about is there's a possibility that you were adding to your cache pressure that you're putting on the CDN for particular pieces of content. If you have an audience that is fragmented across VP9 and H264, you're adding another entry into that cache that you have to make sure has enough traffic for it to stay in there, or you end up going back to an origin and ruining the experience for somebody.
Streaming Media: What about the live side; are you using VP9 there as well?
LaPalomento: We're not using VPN today on live. It's strictly H264. And I think that that's something we definitely are interested in exploring, but we haven't really ruled that out yet.
Streaming Media: Are you still doing capped CRF for both on the VOD side?
LaPalomento: Yes.
Streaming Media: How's that? How does that work as a per-title technology? Is that effective for you?
LaPalomento: Right now we're generally satisfied with capped CRF and we're not actively looking around for different approaches to that. But it is one of those things where there are lots of interesting advancements going on continuously in the field.
And so if some new technique comes up, we would absolutely be very interested in adopting it. There are some really cool things going on with like neural net auto encoders that are fun to look at. But I think those are, you know, years away before they're competitive with the codecs that are hand developed.
Streaming Media: So, you're using VP9 and x264; are those straight from FFmpeg, or are you using custom versions of those?
LaPalomento: Straight from FFmpeg.
Streaming Media: Did you look at any of the custom versions?
LaPalomento: If we have it has been a long time since we actually looked into that. We don't have any recent experience looking at the custom encoders.
Streaming Media: At a high level, what's your sense of the savings that VP9 delivers over H.264?
LaPalomento: Hmm. I wouldn't want to throw a number out without double checking it. And so, I don't have that off hand. So maybe I should follow up with you about that one.
Note to reader: LaPalomento sent the following information after the discussion.
Data rate savings for H264 to VP9 by video width in the JW Player collection. JW Player is using libvpx for VP9 which has been optimized for 480p content and probably explains the higher savings at lower resolutions we see here.
Streaming Media: Which platforms are you delivering VP9 to? Is that going to Android and Chrome? Or just to browsers?
LaPalomento: Well, yes, it's a little bit because there's it's a customer opt in. They have a little bit of control over where that gets deployed. But yeah, it's Android and Chrome are going to be the lion's share, if not all of the VP9 traffic.
Streaming Media: So, you're not using HEVC. What does that tell us about your customer base? I typically associate HEVC with HDR and high-end content. Where do you see that going for you in in the very short term? In the medium term?
LaPalomento: So I would say that we definitely have a large proportion of viewership, which is on websites and in mobile devices and on those platforms that often, you know, the bandwidth savings that you really get out of HEVC aren't as material. So it doesn't have the same kind of cost benefit.
Around the globe, people are looking a lot more into DTV and OTT experiences. And that's where I would expect we're going to see the demand for it. But as of right now, most of our customers have been getting by with H264 on those platforms.
Streaming Media: OK, now, everybody is talking about AV1. How do you approach looking at a new codec at JW?
LaPalomento: Yeah, well, from the technology implementation side of things, as long as there is a reasonably production ready version of the encoder, it's not too hard to slot it into our existing transcoding infrastructure. So that's the nice thing about way this technology landscape is working out.
What tends to push us to evaluate a codec is when we see that the platform and device reach is getting to the point where we feel like we can deploy to a good percentage of our customers. So, I would suspect the way that we're probably going approach AV1 is similar to how we did VP9; we'll make it available to customers who choose to use it.
Over time, our objective at JW Player, is that none of our customers have to worry about what codec they're using on to what devices, into what platforms. And so, in 2022, my guess is that we will be moving VP9 into the space where it will be more automatically configured and available for our customers on their content, where that makes sense.
And AV1 is trailing further behind that. It's just going to take a little bit longer before the cost benefit equation kind of plays out.
Streaming Media: How does that work for codecs like VVC which, all things considered, probably isn't going to be playing in Chrome any time soon.
LaPalomento: For that one, it's unlikely that we would take a look at it for quite a while. Moving into the browser is a big milestone for codecs. And being available on mobile devices is a big one for us. So those two things are kind of big triggers for us to start looking at it and seeing if it makes sense.
Streaming Media: OK, so switching gears, what are you using on the ABR format side? You mentioned HLS before; are you using DASH anywhere?
LaPalomento: Yeah, we do have DASH and HLS available.
Streaming Media: Are you using dynamic or static packaging?
LaPalomento: We use dynamic packaging, at least swapping between TS and fragmented MP4 for DASH and HLS. Now HLS does have the fragmented MP4 option, but it's got more limited compatibility and tends not to be worth it right now to worry about the packaging.
Streaming Media: What's the cost benefit of dynamic vs. static?
LaPalomento: At this day and age, I feel like you need to have a dynamic packaging option of some sort. When you think about DRMing content, you think about the different packaging formats and just being able to adapt to changing device compatibility and different platforms.
Even if we felt really confident that we needed to just have an MPEG TS and fragmented MP4 version of a file, we can store that and be static. As time goes by, that changes. And the dynamic packager lets us adjust a whole bunch of those features of the video delivery to whatever's happening in the technology ecosystem without having to go through and do a massive upgrade, even if it's just like a remuxing of all the content that we have in storage, which is significant.
Streaming Media: Do you see CMAF changing that anytime soon or is that just another ABR format?
LaPalomento: So, I like CMAF a lot. For me it is about minimizing cache fragmentation in the CDN. But still, the capability to do some light manipulation on the content before it goes out to the viewers is incredibly powerful. So even if we were using CMAF and not doing packaging on the fly, having the ability to, for instance, DRM encrypt on the fly, is a powerful thing to have in your arsenal.
Streaming Media: What percent of your content is going out with DRM?
LaPalomento: I don't have that number off hand, but I would guess it's not the majority. I would guess more probably like down towards 20 percent or something like that. It's less than half, let's put it that way.
Streaming Media: What about advertising insertion? And how much does that complicate your live and VOD distributions.
LaPalomento: Advertising does complicate what we do. I think the one nice thing about the way that the ecosystem has moved, particularly with serverside ad insertion, is that you do have a reasonable amount of flexibility in your codec choice as you move from content into ad and vice versa.
On most platforms, you can't switch to VP9 in the back and to H264 in the middle of the stream. But as long as you're consistent with one codec throughout the presentation, you're fine. That's another reason why H264 is going to be hard to unseat. Because you also have to get the ad ecosystem to move along with you if you want to do server-side ad insertion. And the customers who are doing that high quality content that really benefit from VP9 are also pretty interested in doing things like ad insertion.
Streaming Media: OK, listen, that's all I got. I really appreciate you sharing these details with our readers.
LaPalomento: I really enjoyed this conversation. It is a lot of fun to see new codec technology rolling out in the wild. And I appreciate getting a chance to talk about it.
Related Articles
Twitch Senior Video Software Engineer Nikhil Purushe explains how Twitch has approached packaging and delivery following their migration to VP9 in this clip from his Video Engineering Summit presentation at Streaming Media East 2019.
11 Oct 2019
Twitch uses a head-and-tail encoding strategy where popular content (the head) is encoded one way and less popular content (the tail) another. In this interview, a Twitch engineer explains what's on the video gaming powerhouse's five-year roadmap.
23 Apr 2019
The licenses cover devices such as smartphones, computers, TVs, set-top boxes, and graphics cards, but not encoded content—at least for now.
27 Mar 2019