Martin Tobias' startup squeezes out small files on a grand scale.
In 1996, Martin Tobias was living a double life. During the week, he toiled as a mild-mannered Microsoft marketing exec managing online software sales. But on weekends, he revved up his Harley-Davidson Fat Boy and thundered down the highway with a motley assortment of Seattle-area riders to work on Biker Dreams, a brash indie documentary that featured Tobias and covered the annual Harley convention in Sturgis, South Dakota.
"When we finished the film, we thought we should have a Web page and put a trailer on it," recalls Tobias. But he knew from his experience at Microsoft that a movie trailer on the Net would exact a heavy toll in download time from anyone who wanted to see it. Using even the most efficient video compression/decompression, or codec, technology and the fastest modem available at the time, a three-minute trailer would have taken over an hour to download.
Tobias had friends at RealNetworks, an Internet startup that had made a splash in 1995 with the first streaming-audio codec, RealAudio, which could play sound files over dialup Net connections in real time. The company, he learned, was working on a similar codec for video.
"At that time, I was wondering what I wanted to do with my life," Tobias recalls. "Actually, riding my Harley around was looking pretty attractive." Still, he thought, maybe there was a business in this. RealNetworks would supply the technology, telcos would supply the bandwidth, and media outlets would broker the content. But who would encode the thousands of video and audio titles into the necessary compressed formats? There would be no media on the Web, Tobias realized, without someone to digitize and encode it into files small enough to make downloading quick and easy over a Net connection at any speed. At that point, the entrepreneur in him shoved the biker into a sidecar, and Tobias set out to build the ultimate digital-media service bureau, encoding.com.
Founded in July 1997 and still privately held, Tobias' Seattle-based company booked twice as many orders in the first quarter of 1999 as it did in all of 1998. The current client roster includes major media companies such as Sony, Warner Bros., and BMG, as well as Fortune 1000 companies like Cisco and new media ventures like Payperview.com. According to encoding.com marketing director Todd Sawicki, whose official title is Minister of the Medium ("the message is the medium," he quips), the company compresses 500,000 minutes of audio and video every quarter and can read 25 tape formats (everything from cassette to Digital Betacam) and suck data off 1,000 CDs at once, processing the contents into nearly 30 compressed formats.
Although streaming and downloading are often mistaken as interchangeable terms, in fact they're two very different things. Streaming gives Net users instant gratification but relatively poor playback quality (at least for video), and it does not capture the content. Downloading, which involves moving an entire file before any part of it is accessible, offers better quality and - fueling the piracy scare and a host of copyright concerns - deposits the content on your hard drive. Once it's there, you can do pretty much whatever you want with it.
Many encoding.com clients request a streaming format (such as RealMedia), a downloadable format (usually MP3), and a secure downloadable format (such as Liquid Audio or a2b). And almost all clients want files encoded at several data rates to accommodate connections ranging from 28.8 Kbps to T1. Moreover, each encoded file might be accompanied by a number of pieces of related information, or metadata, meshed directly with the file or stored in an accompanying database.
"Our biggest project to date, for one of the major record labels, started with 25,000 CDs," encoding.com's Sawicki reports. "They ended up as 600,000 audio files, as well as metadata such as the title, author, artist, copyright notice, and CD-cover image for each file."
Nearly 100 employees now work inside the company's $2.5 million facility, a warehouse-style space situated in a former church whose facade has been painted black and festooned with a giant neon-purple logo. Black snakes of networking cable crawl across the 18-foot ceiling, dangling down to connect to each desk. At the center of the space stands a line of two dozen 7-foot-tall hardware racks stacked with video and audio decks, processing equipment, and loads of computer horsepower for encoding immense amounts of data simultaneously. But before a file is ready for encoding, it must pass through 1 of 15 high-powered audio/video-capture workstations that sit atop makeshift desks made of doors and filing cabinets.
A compressed signal can be no better than the original, and a bad original - especially one that includes noise - gums up compression algorithms that work by weeding out redundancies (such as the unchanging background behind a talking head). Because its customers show up with everything from Hollywood reels to corporate presentations shot with a camcorder, encoding.com concentrates on optimizing quality before encoding. At one of the workstations, an engineer can scrutinize the original signal for anomalies that might interfere with compression and correct them before putting the squeeze on.
But what really sets encoding.com apart is automation: Once a master audio or video file is optimized for encoding, machines take care of the rest. From the capture workstation, a master and its metadata - including a work order that specifies the desired encoding formats, data rates, and so on - travel via gigabit Ethernet to a "rather big file server," an SGI Origin 2000 with a disk array that holds a whopping 5 terabytes (or 5 million megabytes). The server commands a massive amount of raw computing horsepower: sixty dual-Pentium 450-MHz machines, each equipped with 256 Mbytes of RAM, and more than 20 G3 400s - and every one is outfitted with customized software that can encode the full range of formats encoding.com supports.
Reading the work order, the server doles out the master to successive dual-Pentium or G3 machines as they become available. As each machine finishes its task, it sends its work back to the server and is ready to crunch another file.
Finally, encoding.com ties a bow on the process with a stroke of Jetsons-esque genius: When an encoded file reaches the server, the server can publish it directly to a client's Web site via FTP without requiring so much as the press of a button by a lowly keyboard jockey. At the same time, the server can deliver metadata about each file directly to the client's database. For a client like Licensemusic.com, which provides prerecorded music to video producers and the like, the benefits are manifold.
First, no labor is required to keep track of the thousands of files that are created every week. There's no possibility that files will be misplaced or misnamed, and relationships between different encoded versions of the same original file are tracked in a database. Moreover, encoded files are available to Licensemusic.com's customers immediately, which means they can start generating revenue virtually from the moment they're created.
These automated activities are controlled by metadata, the incidental information that travels along with the actual content of a file. The nifty thing about metadata, Sawicki points out, is that it can consist of anything content providers might find helpful. "You can put an encryption key in there," he says, "or other controls or business rules. You can put in something that says, 'Make sure the file plays only 10,000 times, and after that delete it.'"
Now, encoding.com is preparing for the day when all broadcasting will be done over the Internet. "The bet we're making - the bet that everyone on the Internet is making - is that TCP/IP will replace all other networks, and that we're all going to have a lot of bandwidth to play with," Tobias says. "When that happens, the experience we have now is just going to get better and better."
PLUS
