3:33 – Eric: Start off by thanking Limelight! “It was a pleasure to stand on your shoulders for the Olympics.”
3:34 – Bhavesh: “They rocked!”
3:34 – Eric: The challenge from NBC: Deliver every minute of every Olympic sport to the web. Where possible, deliver HD quality, break the mold on user experience, and leverage existing assets as much as possible.
3:35 – Eric: At the Microsoft MIX conference, first demo of Olympics. Perkins Miller said “This project terrifies me.” Made MS realize the gravity of the project, how they were changing the face of online sports coverage.
3:37 – Eric: The workflow – 2,200 hours covering 25 sports with up to 35 simultaneous live feeds. 19 days of live production. 5000 videos to be produced. And make it redundant.
3:38 – Eric: We introduced a concept of content packages, both LIVE, REWIND, HIGHLIGHTS, and ENCORES.
3:39 – Bhavesh: We were able to take clips that didn’t make the live TV broadcast, because of time or cutting room floor.
3:40 – Eric: (showing demo of a highlight clip – opening ceremonies). Depending on PC we could serve up to 1.5Mbps stream. Smooth scrolling through the video. Multibitrate, Picture in picture, swap a windowed picture for the main viewing area. Demoing on a Mac.
3:41 – Eric: So what happened? How did we achieve these goals?
3:42 – Eric: First, partnership between MSN and NBC to drive eyeballs. MSN ran an Olympics module on top third on site for entire length of Olympics.
3:43 – Eric: What happened? 1.3 billion page views, 50 million uinique visitors, 70 million videos watched, 27 minutes of viewing per video (!!!)
3:43 – Eric: 7 of 10 European countried delivered Olympics content through MS Silverlight. China delivered Olympics on Windows Media Infrastructure.
3:44 – Eric. More stats – 5,000 unique clips viewed per day during final week, 600 million minutes of video, 130,000 peak streams. 3.4 petabytes of video delivered. “It worked flawlessly.”
3:45 – Bhavesh: While the peak streams might not bea big number, remember that there was continuous streams for every day of the the entire event.
3:46 – Eric: Showing a timeline that begins in June 2007 to get to ‘go live’ in August 2008.
3:47 – Eric: Slide showing 10 key partners. Our common goal was to make the Olympics as amazing as possible. We had a “natural attraction” to make it something extraordinary.
3:49 – Bhavesh: Showing the “scary diagram” — the video workflow. Huge eye chart with lots of lines and dependencies, shows how complex it was. We were generating close to 18 gigs of content per hour at the highest point of creation. Now showing a planning chart, showing the amount of content being created for every hour of every day of the Olympics.
3:52 – Bhavesh: We only had a 17×14 room in the broadcast center, which helped us figure how many servers (used processors – Penryn) which limited amount of on-site processing to create the content. Build servers in US, shipped them to Beijing, took two months to arrive.
3:54 – Bhavesh: Back to the scary diagram. Used to plan out which systems were going to be communicating to what – and how much bandwidth they need to talk to each other. Problem we had to overcome was to get all the disparate systems to talk to each other, so that as things change the systems can adapt in real time. We had to come up with a messaging system.
3:56 – Bhavesh: Used MS SharePoint to track clips. MS predefined the types of clips that would be created (“On Day 4 we will have a top 5 goals, a best of boxing, etc.”). Outputted XML which was sent to each partner’s system so it knew what clip it was getting and what to do with it. Even extended to cable ops, mobile operators. Metadata was the key – provided a common messgaing schema for video identification, scheduling, command/control, delivery.
3:58 – Eric: The metadata schema, and doing the planning upfront, let everyone “snap to” that metadata to understand the workflow.
3:59 – Bhavesh:One of the things it allowed us to do was create a “highlights factory.” We took over the SNL Studio in 30 Rock. We had a separate team that would create metadata for every scene, and then students create highlight clips, by going through that metadata. And you’ve got to bve able to create clips very quickly for the web. Not the luxury of time.
4:01 – Eric: Metadata also let you contextualize the content, and not just by time. Its one thing to do voice recognition, but we were able to humanize the video. Were able to get soccer highlights up on the web in 12 minutes to be viewed.
4:02 – Bhavesh: On to the player….both WIndows Media and Silverlight. Sme look and feel in both players, but Silverlight gave you large-screen mode w/16:9 view. Multibitrate experience showed you the best quality video that you could support. Main bitrate at 650 Kbps, but we went up to 1500 Kbps if the user could support it. Could have gone higher, but we erred on the side of caution.
4:04 – Bhavesh: Lessons learned: Reduce complexity via common schema. Even when traffic volume decreased, unique videos viewed grew. The long-tail lesson was there – people really wanted to see all that was out there.
4:05 – Eric: Received emails from viewers. “I’ve never been able to watch equestrian before.”
4:06 – Eric: Also learned the long-tail hides issues. We initially discovered we didn’t have a real-time monitoring solution to look holistically at the entire network condition, consumer connectivity, bad encodes, etc.
4:07 – Eric: Next, unified telemetry and monitoring solutions are imperative. Unified news ultimately led us to make better decisions on workflow, ad inventory. Industry needs to mature from a reporting perspective.
4:08 – Eric: SInce we concentrated on the workflow, we left some on the table. Could have pushed to 1.2, 1.5, 1.7 megabits. Next time, two years, who knows…
4:09 – Eric: These types of projects require a level of intimacy with all developers and partners. Rock solid solution development happens when you meet face-to-face. It made the world of difference. Build in some adidtional time in your SOWs for scrumming – 7 to 10 days should be the longest you go without having some type of build or check-in. It made a world of difference.
4:12 – Eric: In the end, its complicated. Its not just content, its software (to build the workflow and create the consumer experience.). Silverlight was the galanizer – for platform, user experience, delivery, and consumer network.
4:13 – Eric: Why work with Microsoft? We have a rock solid plaform for building rich media experiences. We are “dead set” on innovating to meet the needs of content publishers. But the platform isn’t all. Its about the partner ecosystem (slide with logos). Also 3.5 million .NET developers.
4:15 – Eric: This type of work is in our DNA. Microsoft is in the business to build platforms for people to build great apps on.
4:17 – Time for Q+A.
4:18 – How close did you come to meeting or exceeding your goals for network or storage capacity?
4:18 – Bhavesh: We created more content than we anticpated, some events ran longer, etc. For the most part, we were right on the dot for the video content being created. We were under requirements for power.
4:19 – Eric: On delivery, we made sure that we were ready to serve every customer that wanted to see it.
4:19 – Bhavesh: On one day, we have 113,000 requests per second at our edge. Now looking back, it was fine. But we didn’t know if there was going to be a protest.
4:20 -What was the cost, and did NBC recover their costs through advertising sales?
4:20 – Eric: Cost for development was orders of magnitude over what you would expect for the web. On revenue, we had a mutually beneficial relationship in terms of traffic and monetization. But because this is a partnership I have to respect all parties involved.
4:22 – You had people entering qualitative metadata – what was the editorial policy to determine what was good?
4:23 – Bhavesh: We had editorial staff that filtered everything. Created a process right from the beginning to do this.
4:24 – Thank you!