• Blogs
• Forums
• Tradeshows
• Partners
• White Papers
• Events

On-Demand Computing

Nov 1, 2004 12:00 PM, By Dan Ochiva

Can You Benefit from Renting Rendering?

Sidebars
Hear Comes Sun,Again
For the Little Guy

If you're running an animation or effects facility today, the success of animated images is both a boon and a challenge. Year after year, audiences for movies and commercials grow more sophisticated, pushing creatives to better their skills and deliver unique, new looks. At the same time, animators also want to grow their imaginations, pushing the technology to deliver more realistic fur and skin, for example. Meanwhile, there's always a new special effect to be designed, and editors work to create ever-heftier composites ranging from 100 layers and more.

All good. Now it's time to render. But how?

With the continuous drop in computer technology prices, it's possible for many facilities to build their own render farms, but it's still a huge expense to buy, build, and maintain a massive one. Even though it might be used 24/7 for days and weeks on end, who can afford to then let it lie dormant until the next big job comes along?

“It's constantly a challenge,” says Steve Canepa, vice president of IBM's Global Media & Entertainment Industry division. “It's a difficult business decision to buy equipment and have it on a depreciation schedule where you're trying to refresh the technology on a fairly quick cycle, so that you can stay on the leading edge of technology. The issue for the larger companies isn't just the challenge of building and managing their own [render farm]. Their problem is looking for a more efficient model to stay current on technology and to buy resources when they need them.”

Smaller companies face another problem, Canepa says. Boutiques want to create that same level of craftsmanship as the big houses, but don't have the capital to invest. If you're starting out with only a single film or large commercial in the pipeline, it's just not feasible to sink capital into lots of hardware.

So what's the solution? Some think it's on-demand computing, or utility computing — compute power as another on-tap service like electricity. While it's been discussed for decades, the concept is now coming to market. Major computer vendors have already staked out the DCC space as a prime market to test both the business model and the technology. Over the past 18 months, HP and IBM have developed new business units that employ huge render farms consisting of thousands of microprocessors, all of which can be tapped into via high-speed links and sold as a service. DreamWorks and Threshold Digital Research Labs are applying this new business model.

The History

Today's on-demand computing models build on similar attempts of years past, some of which succeeded, others which have failed. Over the past decade, for example, web-based application service providers (ASPs), managed hosting facilities, and on-demand technical computing helped to establish the concept of compute power on tap as well as “renting” software. In some ways similar to the established outsourcing of back office tasks, businesses began to question why they should buy and maintain complex server arrays, when you can leave it to a third party and save that capital expense and troublesome management.

In 2000, Wamnet, already in the business of high-speed transmission and storage of data, unveiled render on demand (ROD). Wamnet installed SGI workstations in a few test facilities, along with a large installation of expensive SGI gear at the head end to handle rendering. But the new business model quickly ran into problems, according to Tom Moran, Wamnet's director of media and entertainment. Owning the rendering hardware, for example, proved a challenge to a company that based its business as a provider of managed services. “[The hardware manufacturers] don't have to pay a dime for the hardware up front to somebody else,” says Moran. “Because we're not a hardware manufacturer, we would. And that makes the model much less attractive to us since there's a significant capital expense involved.”

Furthermore, Wamnet's business model involved leasing — not owning — high-speed lines from telcos. “The way the telco industry works is that typically, in order to get a decent price, you have to sign a one-year contract for a circuit,” says Moran. “Quite often, what [the animation houses] needed was a true on-demand capacity, where as they were finishing a film and were in a mad dash to the release stage, they suddenly need 10 times the rendering capacity than they're willing to buy hardware to support.”

With up to 60 days needed by Wamnet to provision that high-speed line, the project would be over. So it was great for large productions that could plan far in advance, but it didn't make sense for others who needed the service as jobs rolled in.

Within two years of its launch, Wamnet closed down its rendering service. Would they consider it again? The company, which was acquired by Savvis Communications in 2003, could use its parent's secure, private worldwide network. Currently, Wamnet offers encoding and transcoding services, but the ROD model still wouldn't work, says Moran, unless facilities installed dedicated bandwidth and invested in other services — such as digital dailies — to make such investments worthwhile.

Next up: Intel. In 2001, the world's largest microprocessor manufacturer launched Screamline, a rendering service that was a side business for the company's Online Services initiative, which focused on web hosting.

Screamline charged for rendering by the frame, with rates that ran from 17 cents to $17 per frame. Interestingly, the service was priced based on the complexity of each frame, such as the number of objects, polygons, and light sources in each. Such a complex pricing strategy alone wouldn't appeal to studios that just wanted a job done quickly with a flat fee structure they could work out up front.

Intel also realized that having a go at web services meant that what you were really doing was business consulting, not just building massive server farms. Creating a large consulting service is something companies like IBM make billions doing, but it wasn't a realistic business model for a chipmaker, according to Intel spokesman Bill Calder.

By 2002, Intel was out of the web services business. But, says Calder, Intel likes to think it's in the business even more than before. “We do that by making sure Intel silicon is everywhere, in every rendering farm all over the world,” says Calder. “Watch this space, though. I think we'll probably have some announcements over the next several months, where you'll see some of the big players adopting Xeons with 64-bit extensions and/or a mix of Itanium. And that's for rendering in the Hollywood space.”

On-Demand Today

The Intel and Wamnet outcomes didn't spell the end of on-demand computing for the DCC industries, of course. With Shrek 2, for example, DreamWorks and HP worked together to prove the concept's feasibility. In October 2003, DreamWorks committed to HP's Utility Rendering Service (URS) for the feature animation. “We ended up using about one million CPU hours of rendering on Shrek 2 from URS,” says Ed Leonard, CTO of Glendale, Calif.-based DreamWorks Animation.

While Leonard estimates that the film required a total of some 10 million CPU hours — the rest was handled by DreamWorks' own render farm — HP's service was still important. “Utility computing is interesting to us,” says Leonard. “With the nature of these films, there tends to be a spike at the end. Having to build out the infrastructure to meet that spike really didn't make sense.” By relying on URS to take over when DreamWorks hit its peak needs, the company only had to build out its own infrastructure to handle its average capacity.

So how did DreamWorks make it work? It first installed what can be the most expensive part in setting up such a system: a secure, gigabit, fiber-optic link, that pricey “last mile” to their telco, needed to reach HP Laboratory's Palo Alto, Calif.-based URS facility. At HP's end, DreamWorks simply installed its own rendering software onto the compute farm, which ran Red Hat Linux on HP's ProLiant DL360 servers.

That's one thing to watch out for. Rendering software can create a significant economic limit, since most standard render licenses charge for each processor employed. Using HP's 1,000-processor URS would be prohibitive, taking away all the cost advantages. That's one reason canny studios roll their own, including DreamWorks. Meanwhile, Threshold uses NewTek's LightWave to render because it includes a no-charge rendering license.

Security, of course, is a big concern. “Shrek 2 is our crown jewel,” says Ed Leonard. “It was hard to come up with a security model we trusted. But ultimately, [HP] convinced us that they had a very secure model. They were very willing to listen to us to put into place the sort of monitoring, metrics, and security measures we needed.”

From the on-demand supplier, it's not only a matter of keeping one company's work secure. As more facilities begin to use a service, security becomes much more complex. A good on-demand provider needs to develop more than just a few passwords to protect priceless intellectual property. “When you're talking about core data, like the frames of a movie, you start worrying about how secure is that installation,” says Gene Becker, director of Utility Services at HP Labs. Providing a set of secure mechanisms for sharing those resources is a big deal. “That technology is just starting to exist, and it didn't before,” says Becker.

Anyone wanting to engage an on-demand rendering service, it turns out, also needs to look closely at the service provider's entire infrastructure. While buying a service might appear simple — after all, you just want to render and store your frames — it isn't something to leap into.

Start the process by looking at your own workflow. “I don't care about boxes, anybody can sell you boxes,” says George Johnsen, chief animation and technology officer at Threshold. “What I needed was how to work faster, and IBM was the only company that came to us and said ‘You know, we have all this hardware, but we really want to understand your business process. What do you guys do?’” (Santa Monica, Calif.-based Threshold is nearing completion of Foodfight!, its first full-length digitally animated feature film. See the February 2003 issue of Millimeter for more information.)

The solution, says Johnsen, isn't just on-demand computing. “It's how does your network work, where are your bottlenecks, what's your process, when do you have times when you need more horsepower than you've got? Those kinds of questions are hugely important in designing these systems.”

Threshold, Johnsen found out, was trying to perform traditional animation on hardware and software designed for individuals, rather than for workgroups. While the company now has a mix of IBM xSeries servers, BladeCenter servers, and IntelliStation workstations, Threshold didn't stop after it bought the hardware. “We're working with [IBM] on four or five different levels, including storage solutions, global services, websphere [business integration software and services], and DB2, their databasing program, which has a very sizable visual component,” he says.

An easy-to-use database turns out to be a key issue. Threshold's animators use IBM's DB2 to track file versions so that they work on only the most up-to-date files, not worrying that they've spent a day with a file several versions old.

The Bottom Line

Finally, though, does on-demand rendering really save enough time and money? While it's hard to measure, Johnsen says, over the course of a year Threshold accomplishes about 25 percent more work. “But if you really look at how you do stuff, the thing that you are actually able to do is to chase the dream, to come up with actual, real images, of making the fake stuff real.”

DreamWorks' Leonard concurs on the benefits of on-demand; it's not so much about the money savings, but what the artists gain. “Ultimately, what it provided us was the ability to make much more creative changes much later in our production,” says Leonard. “It also allowed us to make much more visually complex and interesting work in Shrek 2. I'm not sure if it's much of a cost savings in the long run as much as an opportunity to be responsive, to be dynamic in your business judgment.”

While this is fine for big enterprises like DreamWorks and Threshold, what about the many small facilities, boutiques, or even individuals without deep pockets? Is on demand beyond their reach?

Not so, says HP's Becker. “The small studios and creative shops don't have the option because of capital costs of going out and buying a thousand servers for a render farm,” he says. “The idea that they could tap into what to them seems like an infinite number of resources for just a very short time is actually a very transformational notion.”

According to Becker, ten animators in a garage can now take on a :30 commercial that previously would be beyond their ability to bid for. By quickly accessing 100 or 200 CPUs for rendering, HP would make their RFP competitive.

Canepa agrees. “Smaller companies want to produce the same kind of compelling content as the big studios produce, but they don't want to invest in all that [capital expenditure],” he says. “That's especially true if they're just building their business, when they might just have one film in the pipeline, rather than four or five in the queue.”

One added enticement: while buying that 1000 CPU render farm comes out of a company's bottom line, renting out the same capability allows you to charge it against the production.

---

Sidebar

Here Comes Sun, Again

Sun Microsystems gained bragging rights in 1995 as one of the first vendors to gain an exclusive contract to supply a top animation house — Pixar, in this case — with rendering hardware. Pixar built its farm up to some 117 Sun workstations, using it to complete Toy Story. Soon Toy Story 2, A Bug's Life, and Monsters, Inc. followed. Staying with Sun, Pixar finally built out to 250 servers.

But in 2003, Pixar moved its facility to Linux, trading Sun and its Solaris OS for a RackSaver blade system containing 1024 Xeon processors. (Rack-Saver is now Verari Systems.) But Sun hasn't been sitting still. This past September, the Santa Clara, Calif.-based company unveiled a unique, new capability for its N1 Grid computing program. Users will soon be able to access hundreds to thousands of CPUs with a simple pay-per-use charge of $1 per CPU per hour.

At first, this might seem irrelevant to animators and special effects creators. After all, doesn't Sun's Solaris OS run Houdini and little else in the DCC space? Well, that's changed. Sun has struggled over the past few years, losing market share to products built around the near ubiquitous Intel x86 architecture. In a canny move, Sun recently ported Solaris to x86, opening up a huge new market. Sun also formed an alliance with AMD — they'll both develop Solaris' capabilities to run on 64-bit Opteron CPUs. Now, not only does Solaris' version of Unix run on AMD and Intel CPUs, but Windows and a generic flavor of Linux will soon, too.

While HP and IBM won't put out a price sheet — each job is worked out individually — how can Sun quote a price of $1 per CPU per hour? “Because we own [all of the intellectual property],” says Sun's Peter Ffoulkes. “That allows us to build a huge array, put a price out there, and indemnify any users, since we don't have to pay any fees to Red Hat or anybody else for their OS.”

Ffoulkes, group manager for high performance technical computing marketing at Sun, says that such a standardized pricing scheme will open up utility computing for the media industry, enabling more creativity and further expanding a company's options. “If you can put your software on [our N1 Grid], you can rent it,” he says. “That means a small shop that may have had difficulty in getting into the game before can now use our service for say, a day, rent 100 CPUs at a reasonable cost, and really do something.”

While Sun still has to work out the full features of their new rent-a-CPU business, Ffoulkes says that soon-to-come announcements will include partnerships that will be very interesting to the DCC space.
— DO

---

Sidebar

For the Little Guy

New techniques now enable do-it-yourselfers to take even better advantage of today's speedy, lower-cost computing power. One answer recently came to market from GridIron Software. The Ottawa-based company's GridIron X-Factor software enables you to create a cost-effective yet high-speed preview and rendering system for Adobe After Effects.

The X-Factor plug-in allows you to build a high-speed array by tapping the unused power of any computer on your network. Employing a technique that's become the latest hot concept — grid computing — X-Factor software parses out work from a central workstation to a mix of whatever CPUs you can connect to a LAN, whether powerful workstations, office computers, or even laptops. The software knows where every frame sent out resides. So even if one computer is twice as fast as the next, those finished frames are immediately placed back together in the correct order and sent to the monitor.

After Effects' preview and rendering operations gain speed in a linear fashion as each additional CPU takes on a share of the job. Scott Bryant, owner and creative director of Santa Monica, Calif.-based Steam — he's also a longtime After Effects user — expressed astonishment at the performance gains.

“A high-definition project that took over 33 minutes to render locally with After Effects was completed in just one minute 54 seconds using a grid of 10 computers,” says Bryant. “The introduction of X-Factor has fundamentally changed our HD workflow.”

Millimeter contributing editor S.D. Katz concurs on the workflow changes the software creates. X-Factor, a Millimeter Pick Hit of NAB 2004, speeds RAM previews so that After Effects can move from the back room to the main composite suite, says Katz. Now with clients sitting in the composite suite, this new level of interactivity enables After Effects whizzes to compete with the pricey suites down the block.
— DO

© 2008 Penton Media, Inc.