|
Next 7: Crest Animation Studios
Thursday, June 08, 2006
Crest animation studio is a decade old and has given some
path breaking concepts to the Indian Animation industry, such as Morphing. It
does a lot of 3D animation work for international clients. Creating 3D
animations requires lots of processing power. As the number of such jobs
increased, Crest was under pressure to deliver them on time. Even though they
used multiple state-of-the-art, high end servers, the time taken for production
was very long, taking anywhere between 8 to 24 hours, leading to lots of
unproductive time for the creative workforce.
Business
problem
To render more minutes of animation in lesser time |
IT
solution
High Performance Computing- Render Farmmanaged by application called RenderMAX |
|
Implemented
by
In-house Team |
Technology
platform
Linux and OpenSource |
 |
P Krishna
Prasad Head IT, Crest Animation Studios |
They deployed a high performance cluster, rather a
rendering farm with 35 servers, called Render Units connected together over a
10Gbps network, with an application middleware to convert them in to a single
virtual processing unit. They supported 64-bit computing, which could
address more than 4 GB of both virtual and physical memory. The servers could be
regrouped depending upon the type of animation project. Each cluster had a
master node for job scheduling and storage node for storing the files required
for rendering, as well as the rendered output. To manage the clusters, the team
developed their own system called RenderMAX, which is a combination of scripting
routines, Open Source grid engines, and striped-off Linux kernels. The work
process is easy for the animators. They copy their work files to the storage
node from the central SAN storage and schedule the job using the RenderMAX job
submission web interface.
The deployment has obviously significantly reduced the
rendering time. The team tried ran the linpack benchmark on it, and they got a
whopping speed of about 690GFlops.
Page(s) 1
|
Print
Comment
Email
Digg
Del.icio.us
Reddit
Twitter
