IBC
article here
The dream of every VFX facility is to be able to run dynamic
and cost-effective pipelines tailored to each project. In theory, cloud offers
the most flexible infrastructure to do this but with legacy hardware and
legions of artists tied to workstations it is proving harder in practice to
make the shift. A U.S startup is on the verge of launching a VFX studio as a
service which answers many of these problems.
“Everybody says cloud is expensive and painful but the
reality is that cloud is not the problem. It's a design issue,” says Alex
Hatfield, CEO Juno Innovations
and Juno FX. “If you design it right then cloud is insanely effective.”
He and six colleagues have spent three years developing a
platform that can be run on-prem or in the cloud offering almost instant access
to workstations for freelance VFX artists located anywhere.
“We could theoretically out scale any studio in the world if
we wanted to,” Hatfield claims. “And at a fraction of the cost of traditional
workstations.”
It is running trials with several major studios. The
innovation that has caught their attention is the ability to ‘slice’ GPUs to
allow multiple artists to work on the same machine at the same time.
Modern GPUs play a pivotal role in AI and ML by handling
parallel computations and processing large datasets rapidly. Juno has found a way to adapt the concept to
VFX pipelines so that the same GPU is shared among different workloads.
Hatfield explains, “With the AI boom the major developers
were horizontally scaling out to train their models. While they have huge
budgets they're also trying to squeeze as much as they can out of every single
server. The margins are nowhere near as good in visual effects, but we're
trying to squeeze as much out of each render node as we possibly can, so every
single dollar you spend is actually being used.”
There's an environmental benefit too. “Because you're
squeezing more performance out of the servers you don't need to power nearly as
many. You don't need as much hardware because you get more out of it. The
slicing is what enables density and density enables the ability to scale.
Slicing becomes more and more valuable as your team becomes more active.”
All of this is orchestrated in Kubernetes, the open source
system for automating software deployment, scaling, and management, and hosted
in AWS (although any cloud provider could be used).
“The idea is that if a large scale studio decides to come
and use us they can then use the render operator to essentially go through and
launch render nodes dynamically and slice their servers to get more utilisation
out of the servers that they have, whether on-prem or in the cloud.”
Origins story
Hatfield started in the industry as a gaffer in LA before
joining Digital Domain in Florida as stereoscopic compositor then VFX
compositor. After leaving DD, he worked at a number of studios on the East
Coast which is where his vision for developing a way for an artist to work
securely from home was born.
“I'm originally from South Florida and I found myself
travelling up to Boston or New York City for work, staying there for six months
compositing on commercials, renting a supercheap apartment, and coming home to
stay with my parents in Florida for half a year so I could wakeboard. I did that for three years straight. I wanted
a way that I could avoid extensive travel, enjoy the outdoors life and still do
what I love which is to work on movies.”
Right from the start of his career Hatfield had been both
artist and technician, learning scripting languages like Python on his own
initiative, so he could expand and enhance on existing pipelines.
“I was trying to figure out how I can use the cloud so that
myself and fellow artists can do great work from home.”
An earlier attempt to found his own cloud-based studio in
2013 didn’t work out. “It was clunky, like a glorified Dropbox and we didn’t
have the security that studios required.”
While working at Cinesite including on projects including Spider
Man: No Way Home he started learning Kubernetes and began to wonder if he
could containerise workstations in the cloud. This involves building
self-sufficient software packages that perform consistently, regardless of the
machines they run on.
“Some of my friends said this was impossible. But I was at
least going to try.”
Further clues came from observing the way large banks had
set up IT systems to handle masses of simultaneous transactions. “I realised
that 90% of the infrastructure had already been solved over the past 10 years
by other industries. You don't have to reinvent the wheel. You just need to go
look for inspiration elsewhere.
“For instance, when a massive influx of transactions hits a
bank server and the bank is running on Kubernetes, Kubernetes can react to that
and issue more servers to be able to handle the workload. Then when its
finished it will automatically scale back down.
“Our idea was to create a system with the exact same
mentality, but one that is able to handle a massive amount of render tasks.
Once done, we delete the render nodes out from underneath it. If you looked at
our infrastructure, we look nothing like a VFX studio. We look way more like a
financial institution.”
When the VFX industry was hit by redundancies as a result of
the pandemic, Hatfield and his colleagues resolved to go for the final push.
“Could we build tools and applications to make working in
our industry easier and enjoyable for both studios and artists? Why can't we
just sacrifice the infrastructure instead of the people?”
They prototyped the technology using inexpensive Raspberry
Pi boards, a miniature PC and an old game server. “I thought, I’m never gonna
be able to pay for a cloud provider’s bill so what I have to do is build
everything locally and then hopefully if we get a project we'll push it up to
AWS and see if it works.”
They trialled the home-built server on student film projects
managing to run three compositors working off the same stack. “It was far from
perfect but we proved that it worked and at that point I reached out to AWS to
talk about what we were doing.”
With ECR (Amazon
Elastic Container Registry) every component is containerised allowing rapid
replication to any region in the world. They use EFS (Amazon Elastic File System) for storage
along with Amazon Virtual Private Cloud
and EC2 (Amazon Elastic Compute Cloud).
Juno is also able to rely on Amazon’s security.
“When you bring in Kubernetes the actual infrastructure is
so abstracted that you're basically describing concepts,” he says. “That means we can deploy anywhere, scale to
as many workstations as needed and operate with extreme efficiency. We
literally sync the media back and forth which eliminates a lot of the ingest
and download issues.
“We wanted to do really cool work and we also felt like if
we can build it in such a way where we could scale down just as fast as we
could scale up. That means that we could technically compete and even lead the
rest of the industry in a new way of working, to be able to actually work
remotely and access talent anywhere in the world.”
Prepping to launch
In theory, Juno FX could take on projects of huge scale and
work with artists located anywhere in the world. In practice, the startup is
taking things step by step.
“We could go out and compete for the biggest VFX projects if
we wanted to but we’re not really targeting that right now. We want to grow in
a very controlled manner.”
All development to date has been done by the colleagues in
their free time. “We're trying to get enough funding to be able to hire more
full time developers early 2025. We are actively doing PoC deployments for
on-prem density on rack. When they finish in the spring we'll be able to start
supporting bigger accounts. Plus, we're starting to invest a lot of our effort
into getting business from other industries [including medical research,
AutoCAD and architectural visualisation].
“It’s so hard to get through building a startup to the point where you finally see a workstation turn on, you deliver a frame and you get that first pay check but every second has been worth it. This has been our passion. All my colleagues came to my wedding, so they are literally like family to me. And as a side effect, because we believed in it so much, we literally built what we think is the future of visual effects.”
No comments:
Post a Comment