Novartis Novartis has been a speed stone of great concern to a company. In our minds, Novartis innovation in IT technology integration and technical aspects of large pharmaceutical industry, has been the vanguard of the industry, exemplary.
Five years ago, a public case Novartis and AWS issued deeply shocking to us:
“We completed the equivalent of 39 years of computational chemistry in just under 9 hours.”
- Steve Litster, Ph.D., Global Head of Scientific Computing
Less than nine hours, we completed a stoichiometric equivalent of 39 years.
Five years ago , a lot of people in mind, "Cloud Computing cloud computing" concept has only just budding, and as the century-old pharmaceutical industry giant, has begun using cloud capabilities to energize their core drug development process .
Let's look at this case:
Novartis's NovatisInstitutes for Biomedical Research's (NIBR) departments to set up the original intention is to provide a set of treatment for specific diseases and prevention of drug, thus greatly improving the quality of life.
As of 2013, NIBR scientists in the world with 6000, while engaged in 130 drug development projects, mainly focused on the molecular pathways of different diseases of the public.
Traditionally, an innovative drug discovery and development from the beginning to the final launch, often take 10 years --15 years, during which the cost of consumption of more than $ 2 billion, while the huge risk of failure . To address this challenge, NIBR decided to drastically shorten the development cycle by a new generation of HPC systems.
2013, Novartis launched a compound screening programs for certain common cancers, it is necessary to complete the screening of ten million compounds in a week time. According simple estimation, if you rely on Novartis's own IT environment, they may need to 50000 CPU core, the overall cost could be close to $ 40 million.
Fortunately, the emergence of public cloud services, so this problem solved.
Novartis collaboration and AWS and Cycle Computing (later acquired by Microsoft) together for this project, while running 10,600 SPOT Instance, a total of 87,000 CPU cores, so in a short period of 9 hours to complete this historic feat within, the hardware cost is just over 4000 US dollars !
5 years as fleeting, at the end of 2018 AWS Re: Invent conference, Novartis appeared on the podium.
This time, Novartis they are given further thought to HPC and emerging IT technologies - "Ten Strategies on HPC systems in hybrid architecture environment":
Portability.
The design of all new business processes and back-end infrastructure support, and these should be decoupled from the physical site.
Stone-speed Commentary
Decoupling hardware and software platform is a necessary condition for a new generation of hybrid HPC architecture, but also to avoid the Vendor Lock in reasonable measures, although there will be a certain amount of work, but the flexibility obtained absolutely worthy of pre-pay.
Hybrid.
While using its own data center and cloud services to meet the needs of HPC computing. The choice between the two depends on the preset operating principle (time-to-value, cost, capacity and physical limitations, etc.) systems.
Stone-speed Commentary
HPC hybrid cloud is the trend, on-premise and Cloud platforms have their own advantages, are indispensable. The HPC workload allocation should be policy-driven factors (Policy-driven) and a high degree of autonomy (Autonomous), cost, performance, safety, physical distance can all be considerations.
I do not know Novartis has been able to achieve what extent, it is believed that the general direction is wrong.
Multi-cloud with preferred vendor bias.
If you need to use cloud resources, use your preferred cloud service providers by default. If the other cloud providers in some areas there are obvious advantages, it is possible to simultaneously employ a number of cloud providers.
Stone-speed Commentary
Multi-cloud deployment force will be considered HPC bring more flexibility and greater scale, so that end users can get a better cost structure and cost-effective.
Of course, such as the body mass of Novartis customers, there will be a choice of strategic cooperation cloud provider, may be expected to get better prices and support. For smaller customers, preferred vendor bias will not be necessary for the.
Two centers.
The HPC's own resources and the environment is divided into two clusters: one in the United States and one in Europe.
Stone-speed Commentary
Novartis large fortune, their own IT environment can easily achieve multi-site. But for the average user, it is not easy to implement.
But the idea is worth learning, that is, high availability and redundancy, and computing resources closer to the end user. Emerging "edge computing (Edge Computing)" Maybe we can use the tool.
Lease.
The hardware and software procurement model to adjust for the lease as the core, so that it can more effectively track and adapt to rapidly changing computing needs and IT technology.
Stone-speed Commentary
"Rent" in many cases than the "buy" better, this assertion for the enterprise IT environment, more and more self-evident.
In China, individuals and companies for the pursuit of "assets" has a kind of paranoia, which in fact would greatly reduce their efficiency and flexibility, but also a waste of social resources. You can then think about the essence of this strategy.
Environment isolation.
HPC infrastructure should be independent of the general IT environment, and should be optimized separately for research needs.
Stone-speed Commentary
The vast majority of domestic customers, HPC environments and developed using other enterprise IT applications running on the same infrastructure, which in itself can improve resource utilization, but also makes the isolation and tuning HPC environment becomes very difficult. A practical solution is HPC applications and functions are encapsulated to achieve HPC software-defined architecture.
Stage data.
HPC environments only the intermediate data to be processed (Staged Data). All data should be copied into the HPC system, the calculation is complete, then copied out. Transmission of data replication tools should be as easy to use.
Stone-speed Commentary
This strategy, no HPC years of operating experience, is not to mention out. To make HPC environment has become efficient and easy to manage data lifecycle management work should be done by other independent platform.
Shared cost model.
HPC infrastructure costs should be borne by the HPC team, while end users have the right to their own business needs to be dynamic cost burden.
Stone-speed Commentary
Envy Novartis, can make such "utopian" ideas. On further reflection, however, this strategy in general corporate landing is not impossible.
For the domestic business environment and the company's organizational structure features, we recommend the HPC project, using the department should take on more roles. Because the association HPC and business and R & D is so close, it can be seen as the development of production, rather than the IT platform. Use the specific needs of departments, HPC is responsible for the entire project execution, and assume the budget, IT departments to assist in providing technical support, should be a more realistic and more efficient implementation path.
Enable and empower users.
The establishment of support services to better serve the end-user, so that different skill levels of users can efficiently self-service mode using HPC service.
Stone-speed Commentary
Self-service Self-Service is HPC evolution of thinking in terms of functionality, a good set of HPC platform, user experience level, should effectively lower the threshold for the use of business users and the learning curve should not try to change the way business users daily operations.
Planned recurring reviews.
Regularly review the key elements of these strategies to determine whether adjustments are needed.
Stone-speed Commentary
Foreigners have to admire the way of thinking, it is always with the rigorous logic, any methodology, can form a closed loop self iterative update. Of course, business requirements and technical supply is always dynamic, the future does not rule out the emergence of new technology or application mode (for example, edge computing), so regular reviews are proper meaning.
See, in the past five years, Novartis had more experience and experience in the HPC on HybridCloud this respect, able to give this brilliant conclusion, it is fortunate industry.
-END -