By Karen Moltenbrey
Every vendor in the visual effects and post production industries relies on data storage. However, for those studios working on new media or hybrid projects, which generate far more content in general, they not only need a reliable solution, they need one that can handle terabytes upon terabytes of data.
Here, two companies in the VR space discuss their needs for a storage solution that serve their business requirements.
Located in downtown Los Angeles, Magnopus creates amazing VR and AR experiences. While a fairly new company — it was founded in 2013 — its staff has an extensive history in the VFX and games industries, with Academy Award winners among its founders. So, there is no doubt that the group knows what it takes to create amazing content.
It also knows the necessity of a reliable storage solution and one that can handle the large data generated by an AR or VR project. At Magnopus, the crew uses a custom-built solution leveraging Supermicro architecture. As Magnopus CTO Lap Van Luu points out, they are using an SSG-6048R-E1CR60N 4U chassis that the studio populates with two types of tier storage: the cache read-and-write layer is NVMe, while the second tier is SAS. Both are in a RAID-10 configuration with 1TB of NVMe and 500TB of SAS raw storage.
“This setup allows us to scale to a larger workforce and meet the demands of our artists,” says Luu. “We leverage faster NVMe Flash and larger SAS for the bulk of our storage requirements.”
Before Magnopus, Luu worked at companies with all kinds of storage systems over the past 20 years, including those from NetApp, BlueArc and Isilon, as well as custom builds of ZFS, FreeNAS, Microsoft Windows Storage Spaces and Hadoop configurations. However, since Magnopus opened, it has only switched to a bigger and faster version of its original setup, starting with a custom Supermicro system with 400GB of SSD and 250TB of SAS in the same configuration.
“We went with this configuration because as we were moving more into realtime production than traditional VFX, the need for larger renderfarms and storage IO demands dropped dramatically,” says Luu. “We also knew that we wanted to leverage smart caching due to the cost of Flash storage dropping to a reasonable price point. It was the ideal situation to be in. We were starting a new company with a less-demanding infrastructure with newer technology that was cheaper, faster and better overall.”
Nevertheless, choosing a specific solution was not a decision that was made lightly. “When you move away from your premier storage solution providers, there is always a concern for scalability and reliability. When working in realtime production, the concern to re-render elements wasn’t a factor of hours or days, but rather seconds and minutes. It was important for us to have redundant backups. But for the cost saving on storage, we could easily get mirrored servers and still be saving a significant amount of money.”
Luu knew the studio wanted to leverage Flash caching, so the big question was, How much Flash was necessary to meet the demands of their artists and processing farm? The processing farm was mainly used to generate textures and environments that were imported over to a real-time engine, such as Unity or Unreal Engine. To this end, Magnopus had to find out who offered a solution for caching that was as hands-off as possible and was invisible to all the users. “LSI, now Avago, had a solution with the RAID controller called cachecade, which dealt with all the caching,” he says. “All you had to do was set up some preferences and the RAID controller would take care of the rest.”
However, cachecade had a size limit on the caching layer of 512GB, so the studio had to do some testing to see if it would ever exceed that, and in a rare situation it did, says Luu. “But it was never a worry because behind the flash cache was a 60 SAS drive RAID-10 configuration.”
As Luu explains, when working with VFX, IOPS (IO operations per second) is always the biggest issue due to the heavy demand from certain types of applications. “VFX work and compositing can typically drive any storage solution to a grinding halt when you have a renderfarm taxing the production storage from your artists,” he explains. However, realtime development IO demands are significantly less since the assets are created in a DCC application but imported into a game engine, where processing occurs in realtime and locally. So, storing all those traditional VFX elements are not necessary, and the overall capacity of storage dropped to one-tenth of what was required with VFX, Luu points out.
And since Magnopus has a Flash-based cache layer that is large enough to meet the company’s IO demands, it does not have to leverage localization to reduce the IO demand off the main production server; as a result, the user gets immediate server response. And, it means that all data within the pipeline resides on the company’s main production server — where the company starts and ends any project.
“Magnopus is a content-focused technology company,” Luu says. “All our assets and projects that we create are digital. Storage is extremely important because it is the lifeblood of everything we create. The storage server can be the difference between if a user can focus on creative content creation where the infrastructure is invisible or the frustration of constantly being blocked and delayed by hardware. Enabling everyone to work as efficiently as possible allows for the best results and products for our clients and customers.”
Light Sail VR
Light Sail VR is a Hollywood-based VR boutique that is a pioneer in cinematic virtual reality storytelling. Since its founding three years ago, the studio has been producing a range of interactive, 360- and 180-degree VR content, including original work and branded pieces for Google, ABC, GoPro and Paramount.
Because Light Sail VR is a unique but small company, employees often have to wear a number of hats. For instance, co-founder Robert Watts is executive producer and handles many of the logistical issues. His partner, Matthew Celia, is creative director and handles more of the technical aspects of the business. So when it comes to managing the company’s storage needs, Celia is the guy. And, having a reliable system that keeps things running smoothly is paramount, as he is also juggling shoots and post-production work. No one can afford delays in production and post, but for a small company, it can be especially disastrous.
Light Sail VR does not simply dabble in VR; it is what the company does exclusively. Most of the projects thus far have been live action, though the group started its first game engine work this year. When the studio produced a piece with GoPro in the first year of its founding, it was on a sneakernet of G-Drives from G-Technology, “and I was going crazy!” says Celia. “VR is fantastic, but it’s very data-intensive. You can max out a computer’s processing very easily, and the render times are extraordinarily long. There’s a lot of shots to get through because every shot becomes a visual effects shot with either stitching, rotoscoping or compositing needed.”
He continues: “I told Robert [Watts] we needed to get a shared storage server so if I max out one computer while I’m working, I can just go to another computer and keep working, rather than wait eight to 10 hours for a render to finish.”
Celia had been dialed into the post world for some time. “Before diving into the world of VR, I was a Final Cut guy, and the LumaForge guys and [founder] Sam Mestman were people I always respected in the industry,” he says. So, Celia reached out to them with a cold call and explained that Light Sail VR was doing virtual reality, an uncharted, pioneering new thing, and was going to need a lot of storage — and needed it fast. “I told them, ‘We want to be hooked up to many computers, both Macs and PCs, and don’t want to deal with file structures and those types of things.’”
Celia points out that they are an independent and small boutique, so finding something that was cost effective and reliable was important. LumaForge responded with a solution called Jellyfish Mobile, geared for small teams and on-set work or portable office environments. “I think we got the 30TB NAS server that has four 10Gb Ethernet connections.” That enabled Light Sail VR to hook up the system to all its computers, “and it worked,” he adds. “I could work on one shot, hit render, and go to another computer and continue working on the next shot and hit render, then kind of ping-pong back and forth. It made our lives a lot easier.”
Light Sail VR has since graduated to the larger-capacity Jellyfish Rack system, which is a 160TB solution (expandable up to 1 petabyte).
The storage is located in Light Sail VR’s main office and is hooked up to its computers. The filmmakers shoot in the field and, if on location, download the data to drives, which they transport back to the office and load onto the server. Then, they transcode all the media to DNX. (VR is captured in H.264 format, which is not user friendly for editing due to the high-res frame size.)
Currently, Celia is in New York, having just wrapped the 20th episode of original content for Refinery29, a media company focused on young women that produces editorial and video programming, live events and social, shareable content delivered across major social media platforms, and covers a variety of categories from style to politics and more. Eight of the episodes are currently in various stages of the post pipeline, due to come out later this year. “And having a solid storage server has been a godsend,” Celia says.
The studio backs up locally onto Seagate drives for archival purposes and sometimes employs G-Technology drives for on-set work. “We just got this new G-Tech SSD that’s 2TB. It’s been great for use on set because having an SSD and downloading all the cards while on set makes your wrap process so much faster,” Celia points out.
Lately, Light Sail VR is shooting a lot of VR-180, requiring two 64GB cards per camera — one for the right eye and one for the left eye. But when they are shooting with the Yi Halo next-gen 3D 360-degree Google Jump camera, they use 17 64GB cards. “That’s a lot of data,” says Celia. “You can have a really bad day if you have really bad drives.”
The studio’s previous solution operated via Thunderbolt 1 in a RAID-5. It only worked on a single machine and was not cross-platform. As the studio made the transition over to PC from Mac to take advantage of better hardware capable of supporting VR playback, that solution was just not practical. They also needed a solution that was plug and play, so they could just pop it into a 10Gb Ethernet connection — they did not want fiber, “which can get expensive.”
“I just wanted something very simple that was cross-platform and could handle what we were doing, which is, by the way, 6K or 8K stereo at 60 frames per second – these workloads are larger than most feature films,” Celia says. “So, we needed a lot of storage. We needed it fast. We needed it to be shared.”
However, while Celia searched for a system, one thing became clear to him: The solutions were technical. “It seemed like I would have to be my own IT department.” And, that was just one more hat he did not want to have to wear. “At LumaForge, they are independent filmmakers. They understood what I was trying to do immediately, and were willing to go on that journey with us.”
Say Celia, “I always call hard drives or storage the underwear of the post production world because it’s the thing you hate spending a lot of money on, but you really need it to perform and work.”
Main Image: Magnopus
Karen Moltenbrey is a long-time VFX and post writer.