HOW CLOUD HPC IS REDUCING TIME TO INSIGHTS IN PRECLINICAL RESEARCH
share:
As such, the cloud frees scientists from IT administra-
tion, allowing them to focus on research endpoints
and optimal outcomes. Users can build customized
HPC clusters tailored and sized to their research goals.
Cloud HPC providers have created packages opti-
mized for different use cases. These packages, known
as "instance types," have different computing, graph-
ics, memory and storage capabilities. As molecular
modeling is demanding on the visual capabilities of a
HPC system, its instance has multiple, top-end graph-
ics processing units. Genome sequencing, in contrast,
needs highly parallel compute clusters.
Procuring such customized setups is simple and fast.
Spinning up clusters takes minutes. This is important
to IT teams and preclinical research groups alike.
"Everybody wants to ... reduce the time for deploying
server environments," said Russell Towell, senior solu-
tions specialist at Bristol-Myers Squibb.
Bristol-Myers shortened deployment times by work-
ing with AWS to add to its computing capacity. The
system uses a dedicated, encrypted VPN tunnel and
an Amazon Virtual Private Cloud separate from AWS'
public customers. Server environments are built,
secured and qualified by Bristol-Myers locally before
being loaded into the cloud. These precautions ensure
the system meets Bristol-Myers' standards.
Part of the appeal of cloud HPC to Bristol-Myers
and others is the elimination of capital expenditure.
Researchers only pay for the compute and storage
capacity they use. This flexible, pay-as-you-go struc-
ture has enabled research groups that could not
afford on-premise HPC to access compute capac-
ity. For larger companies, the cloud has changed the
economics of expanding beyond existing on-premise
HPC clusters that are at peak capacity.
Accelerating toward insights
When Novartis wanted to virtually screen 10 million
compounds against a cancer target, it calculated it
needed sustained access to 50,000 compute cores-
-capacity that would cost $40 million to install. The
firm had on-premise HPC, but it was running at full
capacity. Even if Novartis suspended all other uses of
the HPC cluster, it would not have had enough cores.
"We were absolutely dead in the water, stuck and
didn't know what do next," said Steve Litster, global
lead for high performance and scientific computing
at the Novartis Institutes for Biomedical Research.
"We had to create a system that was fast, extremely
secure, inexpensive and easy to use."
Litster contacted AWS to see if it could create a