Sardina Systems: All you need to know about Openstack-Based HPC Data Center
The solution for highly efficient OpenStack-Based HPC Data Center
It is not a myth anymore that OpenStack is the leading open-source cloud platform – fastest-growing, supported by a large community of individuals and enterprises, and creating a standard platform we can all use to build open, highly scalable clouds. More than that, OpenStack represents the best of what cloud technology has to offer.
But how is OpenStack integrating into the HPC business?
OpenStack is the HPC user’s dream providing flexible, dynamically scale up resources, while being instantly available.
From the business model, OpenStack in HPC it is delivered as a service to end users (Service Consumers). You pay for what it is consumed.
Another advantage is that it is allowing you to rapidly construct operating environment and it is giving flexible environments tailored by Consumers for Consumers.
From the financial point of view, it is highly efficient because it is built on a model proven successful for Service Operators and Consumers and it is quicker to build, better and more dynamic while having a lower cost.
Last but not least, OpenStack can support live-migrate away from faulty hardware ensuring application and data integrity.
Keeping in mind all the above, we have created a list of the most frequent challenges to be addressed in the HPC environments, as it follows:
- Dynamically and automatically blending the resource demand
In HPC environments, there are workloads which are integer-instruction dominant, while other workloads are floating-point-instruction dominant. We need a solution to automatically takes this into account in deciding on the optimal blend of VMs across all the physical hosts in the data center.
- Dynamically right-sizing a data center
The number of servers can be dynamically right-sized to meet workload requirements, reducing power consumption and boosting resource utility.
- Eliminate fully allocated idle data center
A common problem in OpenStack cloud is that while resources may be allocated to VMs, but the VMs may be idle, thus resulting in fully allocated but idle data center. Such waste of server resources is not beneficial for the cloud operator.
The solution is to continuously track detailed resource utilization, and understand their low levels of resource utilization and system activity, to dynamically consolidate the idle VMs onto as few physical servers as possible while ensuring their low levels of resource demand remain met. This ensures that there would be no fully allocated idle hosts, eliminating wastage.
- Optimally and rapidly places VMs
The optimal, rapid placement of VMs in a large, dynamically changing data center is a major challenge.
In an HPC environment, this could be particularly attractive if the workflow involves frequent starting and stopping large numbers of VMs, for example, in particle physics.
- Highly available architecture
Downtimes are detrimental to productivity. For a large HPC facility, hardware failure occurs quite frequently (due to MTBF being divided by a large denominator of the number of parts).
- Health diagnosis framework and fault-prevention
Track the health of the host servers and auto-migrate the workload to another host.
- Flexible and highly operable
You need to handle different types of operational automation that might otherwise require operator intervention. This design allows a single operator to manage a data center of thousands of host servers — not a typical in large-scale HPC facilities.
Please check our FishOS Product page here.