Evolving to a virtual private local cloud?
By Beth Marchant
VFX artists, supervisors and technologists have long been on the cutting-edge of evolving post workflows. The networks built to move, manage, iterate, render and put every pixel into one breathtaking final place are the real super heroes here, and as New York’s The Molecule expands to meet the rising demand for prime-time visual effects, it pulls even more power from its evolving storage pipeline in and out of the cloud.
The Molecule CEO/CTO Chris Healer has a fondness for unusual workarounds. While studying film in college, he built a 16mm projector out of Legos and wrote a 3D graphics library for DOS. In his professional life, he swiftly transitioned from Web design to motion capture and 3D animation. He still wears many hats at his now bicoastal VFX and VR facility, The Molecule —which he founded in New York in 2005 — including CEO, CTO, VFX supervisor, designer, software developer and scientist. In those intersecting capacities, Healer has created the company’s renderfarm, developed and automated its workflow, linking and preview tools and designed and built out its cloud-based compositing pipeline.
When the original New York office went into growth mode, Healer (pictured at his new, under-construction facility) turned to GPL Technologies, a VFX and post-focused digital media pipeline and data infrastructure developer, to help him build an entirely new network foundation for the new location the company will move to later this summer. “Up to this point, we’ve had the same system and we’ve asked GPL to come in and help us create a new one from scratch,” he says. “But any time you hire anyone to help with this kind of thing, you’ve really got to do your own research and figure out what makes sense for your artists, your workflows and, ultimately, your bottom line.”
The new facility will start with 65 seats and expand to more than 100 within the next year to 18 months. Current clients include the major networks, Showtime, HBO, AMC, Netflix and director/producer Doug Limon.
Healer’s experience as an artist, developer, supervisor and business owner has given him a seasoned perspective on how to develop VFX pipeline work. “There’s a huge disparity between what the conventional user wants to do, i.e. share data, and the much longer dialog you need to have to build a network. Connecting and sharing data is really just the beginning of a very long story that involves so many other factors: how many things are you connecting to? What type of connection do you have? How far away are you from what you’re connecting to? How much data are you moving, and it is all at once or a continuous stream? Users are so different, too.”
Complicating these questions, he says, are a facility’s willingness to embrace new technology before it’s been vetted in the market. “I generally resist the newest technologies,” he says. “My instinct is that I would prefer an older system that’s been tested for years upon years. You go to NAB and see all kinds of cool stuff that appears to be working the way it should. But it hasn’t been tried in different kinds of circumstances or its being pitched to the broadcast industry and may not work well for VFX.”
Making a Choice
He was convinced by EMC’s Isilon system, based on customer feedback and the hardware has already been delivered to the new office. “We won’t install it until construction is complete, but all the documentation is pointing in the right direction,” he says. “Still, it’s a bit of a risk until we get it up and running.”
Last October, Dell announced it would acquire EMC in a deal that is set to close in mid-July. That should suit The Molecule just fine —most of its artists computers are either Dell or HP running Nvidia graphics.
A traditional mass configuration on a single GigE line can only do up to 100MB per second. “A 10GigE connection running in NFS can, theoretically, do 10 times that,” says Healer. “But 10GigE works slightly differently, like an LA freeway, where you don’t change the speed limit but you change the number of lanes and the on and off ramp lights to keep the traffic flowing. It’s not just a bigger gun for a bigger job, but more complexity in the whole system. Isilon seems to do that very well and it’s why we chose them.”
His company’s fast growth, Healer says, has “presented a lot of philosophical questions about disk and RAID redundancy, for example. If you lose a disk in RAID-5 you’re OK, but if two fail, you’re screwed. Clustered file systems like GlusterFS and OneFS, which Isilon uses, have a lot more redundancy built in so you could lose quite a lot of disks and still be fine. If your number is up and on that unlucky day you lost six disks, then you would have backup. But that still doesn’t answer what happens if you have a fire in your office or, more likely, there’s a fire elsewhere in the building and it causes the sprinklers to go off. Suddenly, the need for off-site storage is very important for us, so that’s where we are pushing into next.”
Healer honed in on several metrics to help him determine the right path. “The solutions we looked at had to have the following: DR, or disaster recovery, replication, scalability, off-site storage, undelete and versioning snapshots. And they don’t exactly overlap. I talked to a guy just the other day at Rsync.net, which does cloud storage of off-site backups (not to be confused with the Unix command, though they are related). That’s the direction we’re headed. But VFX is just such a hard fit for any of these new data centers because they don’t want to accept and sync 10TB of data per day.”
His current goal is simply to sync material between the two offices. “The holy grail of that scenario is that neither office has the definitive master copy of the material and there is a floating cloud copy somewhere out there that both offices are drawing from,” he says. “There’s a process out there called ‘sharding,’ as in a shard of glass, that MongoDB and Scality and other systems use that says that the data is out there everywhere but is physically diverse. It’s local but local against synchronization of its partners. This makes sense, but not if you’re moving terabytes.”
The model Healer is hoping to implement is to “basically offshore the whole company,” he says. “We’ve been working for the past few months with a New York metro startup called Packet which has a really unique concept of a virtual private local cloud. It’s a mouthful but it’s where we need to be.” If The Molecule is doing work in New York City, Healer points out, Packet is close enough that network transmissions are fast enough and “it’s as if the machines were on our local network, which is amazing. It’s huge. It the Amazon cloud data center is 500 miles away from your office, that drastically changes how well you can treat those machines as if they are local. I really like this movement of virtual private local that says, ‘We’re close by, we’re very secure and we have more capacity than individual facilities could ever want.’ But they are off-site and the multiple other companies that use them are in their own discrete containers that never crosses. Plus, you pay per use — basically per hour and per resource. In my ideal future world, we would have some rendering capacity in our office, some other rendering capacity at Packet and off-site storage at Rsync.net. If that works out, we could potentially virtualize the whole workflow and join our New York and LA office and any other satellite office we want to set up in the future.”
The VFX market, especially in New York, has certainly come into its own in recent years. “It’s great to be in an era when nearly every single frame of every single shot of both television and film is touched in some way by visual effects, and budgets are climbing back and the tax credits have brought a lot more VFX artists, companies and projects to town,” Healer says. “But we’re also heading toward a time when the actual brick-and-mortar space of an office may not be as critical as it is now, and that would be a huge boon for the visual effects industry and the resources we provide.”