Video services

Internet Video's Scale Scare

SAN JOSE, Calif. -- While much of the focus is on where to get revenues, Internet TV also faces a question of how to connect live feeds to millions of viewers.

The issue came up at VON this week as the NCAA men's basketball tournament was coming to nearby HP Pavilion. (See VON Madness.) That's fitting, because March Madness was frequently held up at VON as an example of a mass-scale live event.

Major League Baseball Advanced Media LP, which handles baseball's online video and also does consulting jobs, helped stream March Madness video for CBS's site last year, reaching 313,000 users. That's a lot, but it's just a "quaint business" compared with the potential audience of millions that sports leagues hope to capture, says Bob Bowman, MLBAM's CEO.

Speaking at the conference here, Bowman described how MLB has built up experience with live streaming video since 2002. The 20 games provided online that year showed how much harder live video is than video-on-demand.

"I'm happy to say every one of them was an abysmal failure," pockmarked with freeze-ups and dropped connections, Bowman said. "More laptops were thrown, mainly by me."

About 3 million people per day view some form of video on MLB's site, Bowman said. Much of that is video-on-demand -- baseball news shows or highlight clips. To serve live game feeds by the millions is going to require some new tricks.

Among them will be the caching of live streams, which MLBAM expects to start this year. MLBAM is also working with Swarmcast to see if a peer-to-peer (P2P) method might be the answer, Bowman said.

Exhibitors at VON's video pavilion were split on P2P. The more established firms -- like The FeedRoom, which provides video technology for outfits like The New York Times -- say content delivery networks (CDNs) have done just fine so far.

FeedRoom officials say March Madness, in a previous year, is the only time they've had trouble with scale. In that case, the company used multiple CDNs, feeding them in round-robin fashion: Akamai Technologies Inc. (Nasdaq: AKAM) would handle one viewer's feed, Limelight Networks Inc. (Nasdaq: LLNW) the next, and so on, evenly divvying up the traffic.

Vividas, a video supplier similar to The FeedRoom, is considering P2P, but primarily for the cost savings; high-traffic events can still be handled with multiple CDNs, says Iain Molland, Vividas CEO.

"The peer-to-peer is attractive because of the cost model," Molland says, adding that his company is considering the technology. "But we use the CDN because we think it's the best way to do it right now."

And it's worth noting that CDNs have been working on beefing up scale. An Akamai spokesman says his is the sole CDN delivering 56 of this year's March Madness games.

Some companies are pushing P2P connections as the answer. Newcomer Neokast claims it's got a way to stream to "infinite" users via P2P. In the non-HD realm, a company called Network Foundation Technologies LLC makes similar claims and has broadcast live events such as OzzFest -- but mostly for standard-definition feeds that aren't full-screen sized.

Then there's the question of high-definition TV. For a recent report, "DSL Video Bandwidth Crunch," Light Reading Insider polled encoding vendors on the bandwidth needed for online TV, given advances in compression. The consensus is that HD video will be scrunched into feeds as skinny as 4 Mbit/s within two years. That's meant to be good news, but still a long way from the 700 kbit/s many video providers are using.

Table 1: Video Compression Timeline
Improvement Standard Definition High Definition
MPEG-2 3-6 Mbit/s 16-19 Mbit/s
MPEG-4 AVC/H.264 Now 2-3 Mbit/s 6-7 Mbit/s (720p) /
8-10 Mbit/s (1080i)
Future 1.5 Mbit/s 4-6 Mbit/s (720p) /
6-7 Mbit/s (1080i)
Timetable to achieve End 2008 Mid to end 2008
Source: Light Reading Insider

Some providers claim they've got proprietary ways to squash an HD signal even further. Vividas, for instance, says it can transmit HD on a 1-Mbit/s channel.

Bowman says it doesn't matter yet. MLBAM streams at 750 kbit/s and doesn't yet do HD online, even though cable companies have asked about that; Bowman's group just isn't convinced that live HD streaming can work on a large scale.

— Craig Matsumoto, West Coast Editor, Light Reading

Pete Baldwin 12/5/2012 | 3:11:26 PM
re: Internet Video's Scale Scare This is one of those times where bigger entities (Major League Baseball) see a problem looming and all the startups say they've got it licked already.

Things are moving fast enough in this area that it's possible they're both right, but the startups are too untested for the MLBs to trust yet.

Of course, the real test comes when you actually get 1 million people to tune in to some live event all at once. That's gonna be a while.
PO 12/5/2012 | 3:11:24 PM
re: Internet Video's Scale Scare It's certainly an interesting point to raise that the carrier billing model will influence the technology choice.

But ultimately, that is probably true at numerous orders of scale.

As higher-volume events become more common (March Madness, Nascar, MLB, FIFA & EPL, Olympics, etc), we can expect to see some sort of regional distribution networks, to minimize the number of copies of streams which have to travel long distances over the same pipes. (This is comparable to the model which drove Akamai's server farms for static content, but for proxied streams.)

For low-volume events, the cost of managing so many underutilized servers overtakes the cost of bandwidth, and new clients can be fed on a loadsharing basis among a few servers which are not necessarily located in the client's region. This is probably the default model today.

Ultimately, an MSO can partner with the content producer to serve as the distributor for their customer base - and achieve cost advantages, marketing advantages, and revenue advantages for premium content. I suggest this is probably the model for development of the server farms, and a motivating factor for those carriers to keep their billing model in line with this path.

At that point, we're back into all the discussions of the future of IPTV, and the limited value of the closed systems which some operators have proposed and deployed.
ethertype 12/5/2012 | 3:11:24 PM
re: Internet Video's Scale Scare P2P is going to turn out to be even less economic than CDN. Why? You have to push data down to participating user PCs, then back up the skinny upstream. In both cases the amount of data will be measured by the broadband provider and counted against a usage limit.

What limits, you ask? At present, usage limits are vague and rarely enforced in the U.S., and most users don't think they exist, but they do. However, the limits are explicit and excess usage is priced very clearly in places like Belgium and the UK, and (trust me on this one) they will be in most other countries within five years. Once users realize that participating in a P2P network jacks up their usage and their costs, they'll opt for "receive only" and become non-participants for forwarding to other users.

Well, you say, maybe someone will invent a way to pay users for the use of their bandwidth in P2P delivery. OK, could happen. But once P2P is transformed from a freeloading arbitrage play, based on the fiction that broadband is "unlimited", to an explicitly priced delivery service, the economic fundamentals will favor a moderately dense CDN network because it gets stuff delivered more directly and efficiently.
flyingsausage 12/5/2012 | 3:11:16 PM
re: Internet Video's Scale Scare The relatively low upstream bandwidth (compared to downstream) of most Internet Access worldwide is maybe the major issue with P2P architectures.

For P2P to work as a whole, the sum of downloads should be equal or lower than the sum of uploads. In case of low upstream bandwidth access, lets say 128Kbps, watching a 4Mbps HD stream would require 32 seeders for 1 watcher. This will never scale for a mass distribution system.

The evolution towards symmetric broadband access (maybe with fiber to the home) will definitely improve this. Having 50/50Mbps access everywhere would make the p2p video approach to scale.

Other aspects could be the multicast optimization. In order to solve the limited # of p2p seeders, one might wish to enable them to become multicast sources, so one seeder can serve a virtually unlimited # of watchers. However, service providers will never allow such an uncontrolled utilization of their multicast ressources, as it's a major security threat (you can easily imagine a simple virus could virtually charge their networks 100% until all client PCs are cured...).
Sign In