In Broadcast - October 2015 - 71

71

www.inbroadcast.com | Vol: 5 - Issue 10 | October 2015

InTech

Making Movies In The Cloud:
The Next Step For VFX
By Matt Provost, Sr. Systems Engineer at
Avere Systems...

T

he increasing use of visual effects
(VFX) in film, television and
advertising has raised the expectations
of production standards. This has
fuelled a colossal growth in server
power and storage capacity to cater
for the growing rendering demands of
VFX.
Many of the larger studios initially
opted to take the investment hit and
provision their own data centres,
managing the peaks and troughs of
production schedules themselves. The
result was a need to overprovision
for busy times, which meant that a
percentage of the hardware would be
redundant and unused during slack
periods.
There is an obvious attraction to
eliminating unused resources with
cloud services where the number of
servers and the storage capacity can
be scaled up and down according
to need. This can have immediate
financial benefits because the cost
of linking to and using cloud is
counterbalanced by the reduction
in the maintenance overhead of
supporting a local data centre and
the expense of cooling systems and
electrical power, some of which is
consumed just to keep unused systems
on standby.
If these were the only considerations
there would have been a massive
rush to the cloud. This obviously has
not happened and there are two
main reasons. The first is the colossal
amount of data movement required.

The second reason is latency - the
amount of time it takes for electronic
signals to traverse the distance
between the studio and the remote
cloud data centre.
Data moves across a fibre-optic
network at the speed of light but,
as every astronomer knows, this
is not instantaneous and all of the
milliseconds of delay mount up and
can impact the success or failure of a
project.
The main conflict lies in the fact
that the speed and efficiency of a
computer system declines as the
distance between the stored data and
the computer increases. To move all of
the data for a project up into the cloud
would occupy too much bandwidth,
cause delay, and disastrously affect
any other work that was attempting
to run at the same time. A reasonable
alternative is to keep the bulk of
the data in the studio facilities and
transfer batches of data as needed to
the cloud, but this would disrupt other
processes unpredictably and latency
would also become a serious issue.
The answer is to consider a managed
cache, such as the Avere vFXT Edge
filer which is a software version of
Avere's solution running as a cluster
of virtual machines inside the cloud
facility to bring the rendering data
as close as possible to the compute
instances. The vFXT cluster will act
as a cache, connect back to the onpremises storage over the WAN, and
handle all connections from the render

clients locally inside the
cloud platform's lowlatency network.
Using this process
means that files are
cached on-demand on
the cloud platform's high performance
storage, utilising fast solid-state disks
(SSDs). The parallel nature of render
jobs involves a high amount of reuse
of the same input files and tools for
every frame in a shot. Storing this data
locally in the cloud means it can be
served out multiple times to different
render nodes.
Another benefit of cloud is that,
because each shot is rendered frameby-frame, compute access can go
"deep and not wide." This means more
servers can be used because renting
10,000 servers for one hour costs the
same as using 1,000 servers for 10
hours. With a deadline approaching,
being able to scale up quickly to
perform renders in parallel will get the
results back in front of the VFX artists
faster - and this is a critical advantage.
A vFXT cluster reduces the increased
load.
Access to a massive farm of servers
through the cloud using Avere blows
the VFX market wide open to allow
smaller companies that would not
be able to afford huge data centres
to enter the big league. They can
offer their clients more sophisticated
results without having to panic-buy or
rent extra equipment to expand their
on-premises facilities. The amount of

Specialised Storage Infrastructure cont'd...
used only for demanding real-time
operations. This approach keeps
performance
highly
predictable,
reduces overall storage system costs,
and improves manageability.
Non-real-time operations can be
moved to extended online storage,
where the content protection is
highly resilient and maintained for
longer periods of time. Transcoding,
rendering, and archive on ingest and
delivery are handled on extended
online with no impact on online
operations. Long-term projects and
frequently used assets also may be

securely maintained there for rapid
repurposing.
Extended online delivers greater
content protection than traditional
RAID, as well as higher capacity and
lower cost per TB, making it perfect
for non-real-time components of the
workflow.
Furthermore,
extended
online
storage makes the long-term archive
of content and assets much easier.
Instead of data tape - either LTO or
LTFS - being attached directly to the
online storage and possibly having
a negative impact on real-time

performance, tape is attached to
extended online storage. The media
asset manager allows users to set
policies that automatically archive
content.
Of course, any modern workflow
must be ready for the cloud. Dealing
with cloud architectures can be
difficult and confusing, and latencies
can be hours or days. Visibility
and integration into facility-based
workflow is key to a successful cloud
deployment, and this is an area in
which a media-optimised storage
solution excels, making it possible

Avere cloud Compute & Storage diagram

compute power required is determined
by the client's budget, not by the
cash reserves of the VFX company.
This frees up money to invest in more
powerful software for the VFX artists.
Cloud services costs can be massively
reduced where per-minute billing is
available. Render-queue software
can automatically spin servers up as
needed and shut them down as soon as
the queue is empty, creating savings of
up to 70%.
Savings in bandwidth costs are
also realised by using Avere and
typically saves 75% of the Internet
link bandwidth that would normally be
demanded by a rendering system. This
lower demand eliminates the need for
increasing the size of the connection
and reduces ongoing bandwidthrelated costs.
James Blinn, a NASA computer
animation scientist, postulated a
theory that rendering time remains
constant despite technology advances.
This has become known as Blinn's
Law and simply means that artists will
always take advantage of increased
compute power to create more
complex, higher-quality frames, shots,
and movies. Without the cloud and
Avere this would mean ever-increasing
costs for VFX studios.
to extend archive or even workflow
to the cloud. A specialised storage
infrastructure such as this can benefit
any production facility, no matter
the production workflow. Efficiency
and predictability can be improved
immediately through the benefits of
integration and automation, while
new capabilities and opportunities are
opened up for the future.
Enabling staff to be more creative
has never been more important with
so much content out there, so any
solution that frees up their time by
removing operational headaches and
stopping potential problems can only
enhance a business.
http://www.inbroadcast.com http://www.averesystems.com/
Table of Contents for the Digital Edition of In Broadcast - October 2015
