Virtual fabric stitches high-performance computing (HPC) cloud platforms together, providing a cloud experience to IT administrators and end-users with HPC clusters.
HPC clusters, by definition, host demanding simulation and analytics workloads, such as computational fluid dynamics, seismic processing, genome sequencing, fraud detection, and more. CIOs are now expecting HPC to be as easy to use as any information system’s component. That’s a good reason why the HPC community is under pressure to provide a cloud experience to IT administrators and end-users with HPC clusters. HPC clusters should be consistent when deployed on-premises, off-premises and when hosted in a public cloud. HPE provides the right mix of HPC capabilities and solutions that allow you to use the same tools and environment to manage your workloads in the same way irrespective of where it actually runs.
Typically, HPC clusters often utilize high-speed fabrics to interconnect compute nodes. HPE along with service provider partners such as Advania Data Centers provide dedicated HPC cloud infrastructure with Intel Omni-Path Architecture (OPA) 100 Gbps fabric.
Challenges faced by existing HPC clusters
The reality is, for most organizations, HPC capacity requirements are not constant and on occasion more HPC capacity is required than is available from an installed cluster. One way to solve this problem is to upgrade and expand the installed cluster, another way, providing more flexibility is to adopt a Hybrid HPC strategy. This hybrid IT strategy allows bursting to a HPC cloud when required, providing optimal use of current assets while increasing operational agility in adding extra needed capacity.
Did you know?
- HPE and Intel® offer the cloud experience recast as an HPC cluster for balanced performance, lower latency, greater capacity, and improved efficiency
- HPE Hybrid HPC delivers the flexibility to burst to a HPC cloud when required, providing optimal use of current assets while increasing operational agility
- HPE and Intel’s critical platform innovations in memory, storage, and acceleration technologies cater to the complex spectrum of diverse HPC workload requirements
Reinventing HPC with the notion of multi-tenancy
Intel® Omni-Path Architecture (OPA) has the capability to generate so-called virtual fabrics (i.e. vFabrics). This provides traffic isolation, an important security feature to support multi-tenancy in a traditional HPC environment transforming an HPC cluster into a cloud.
The user has the same experience of a standard HPC cluster as they do consuming public cloud resources. Creating a cluster, growing or shrinking your partition based on application needs—it’s like the cloud experience recast as an HPC cluster.
Here’s how it works
Dynamic Fabric Configuration is a feature designed for fabric cluster environments, where it is desirable to change virtual fabrics whilst in progress, such as multi-tenant hybrid IT environments. Intel OPA virtual fabrics can be modified and applied to the fabric without stopping and restarting the Subnet Manager. These changes are also non-disruptive to the cluster. To find out more about dynamic vFabrics, read Section 6.0 Virtual Fabrics of this Intel user guide.
The HPE HPC Infrastructure as a Service (IaaS) leverages innovative foundational technologies from Intel. Built for the continuing convergence of HPC and AI, and supported by a broad software ecosystem, the Intel-based HPC platforms offered by HPE provide balanced HPC system performance, lower latency, greater capacity, and improved efficiency.Intel’s innovative HPC foundation, based on the Intel® Xeon® Scalable processors, includes critical platform innovations in memory, storage, and acceleration technologies to address the complex spectrum of diverse HPC workload requirements. Given the criticality of HPC fabric in scale-out deployments, Intel Omni-Path Architecture provides low-latency interconnect for scalable performance of multi-node environments.
The HPC IaaS environment provides customized tenant clusters tailored to client and application requirements. Service providers can use any HPE server in their environment supported by HPE Insight CMU, Docker Enterprise Edition (Docker EE), and OpenStack.
The HPE HPC IaaS platform is essentially an API engine where the IaaS software can be used directly through the HPE API or integrated with a Cloud Management Platform.
IaaS 1.0 allows tenant provisioning and management using the following steps:
- Tenant creation and termination
- Storage volume creation and mounting
- Compute node addition and removal
Using the tenant you created, you can run your native application workflows interactively or as individual jobs or batches of jobs.
Dynamic Fabric Configuration is the foundation of the HPC IaaS platform developed by HPE, as shown here.
Playing with multi-tenancy and HPC
HPE also has Centers of Excellence (COE) dedicated to HPC cloud to support your company as you start your journey. Within the COEs, HPE offers trials and support for proof of concepts with its ecosystem of partners. HPE also develops in the COEs further advanced mechanisms for multitenancy management and IaaS.
First read about the HPC Cloud in this technical white paper: A Platform for Delivering Superior HPC Cloud Services. Next, explore how HPC cloud can transform your on-premises, off-premises, and hybrid HPC experience today by reading this technical white paper on hybrid HPC cloud.
For more information, get in touch with your local HPE representative today.
Hewlett Packard Enterprise