The current era of computing is driven by data. In this data-fueled economy, businesses and organizations are increasingly applying high-performance computing (HPC) techniques to harness the power of advanced analytics and machine learning.
HPC is the use of “parallel processing” techniques to solve complex computational problems through computer modelling, simulation and analysis. Traditionally, most users of HPC systems have been scientific researchers in academic institutions, engineers and specific government agencies (such as the military). Many businesses are reluctant to commit to capital expenditure for HPC infrastructure and are increasingly turning to cloud because it is scalable, available on demand and quick to deploy. This is leading to a surge in demand in the “HPC in the cloud” market.
HPC in the cloud
The cloud seems is an HPC designer’s dream. It offers instantly available and scalable computing resources and almost unlimited storage at a reasonable cost. However, not all HPC in the cloud solutions are created equal. Let’s take a deeper look at the workload characteristics of HPC applications to get a better understanding of their cloud affinity. HPC applications can be segmented into the following sets:
Processing-sensitive applications, which use highly parallel systems that perform complex algorithmic calculations (single Instruction/multiple data operations) and tend to run close to the “metal.” These applications are designed as a set of non-dependent, run-parallel jobs and can be spawned across a large cluster with almost linear scalability (for example, gaming and digital rendering workloads).
Interconnect-sensitive applications, which require low latency and high throughput interconnections. These applications utilize “user-space” communication pathways, bypassing the OS Kernel.
Input/output-sensitive applications, which are highly transactional in nature. They are designed to process a large volume of queries and require a very fast input/output subsystem (for example, financial workloads).
A “traditional cloud” comprised of single, loosely coupled compute instances (such as an instance of an OS running in a virtualized environment) and storage doesn’t provide the best environment for hosting applications that push on all levels of performance (compute, interconnect and storage). Attempting to run this type of application on a typical cloud will produce subpar performance, often resulting in failures.
SoftLayer also offers GPU engines (from NVIDIA) that work in conjunction with a server’s CPU to accelerate application and processing performance. In high-performance computing, GPUs are better than CPUs alone because they have thousands of small, efficient cores that are designed to process information faster.
SoftLayer can provide a high-speed interconnect option in the form of an InfiniBand interconnect and storage SSDs that can be provisioned to provide a high-performing storage subsystem. With these configurations, SoftLayer provides an exceptional, high-performance system at all levels (compute, interconnect and storage) for hosting HPC workloads.
SoftLayer provides a range of network connectivity options, including direct links and a data transfer service, which are critical elements to deploy HPC in a truly hybrid environment. Clients can have the flexibility to provision their entire HPC cluster in the SoftLayer cloud or provision a part of the cluster on premises and the rest in a public cloud.
With these powerful, flexible and reliable offerings, SoftLayer is at the forefront of delivering HPC in the cloud, making it a simple and cost-effective solution for organizations to meet the high performance computing demands of their businesses.
How will your business benefit from HPC on SoftLayer?
There are several success stories of organizations that have deployed HPC on SoftLayer. To cite a few, MapD, New York University and Artomatix have all experienced significant benefits by using the NVIDIA GPUs on SoftLayer cloud services.
Watch as MapD explains their HPC experience with SoftLayer bare metal servers:
Cloud migration challenges continue to bedevil enterprises, despite the fact that the cloud itself has been around for nearly 20 years. Many enterprises have yet to realize the full promise of an abstract, distributed, federated data environment because migrations are still often so difficult. Nobody embarks upon a cloud migration expecting to fail, of course. […]
The term “cloud first” has different meanings for different organizations, and no two companies’ journeys to a cloud-first position are exactly alike. That’s why outlining a cloud strategy road map, a thorough guide to embracing the cloud, can help organizations prepare for their specific needs. A cloud-first strategy shapes how enterprises handle both technology and […]
As organizations undergo digital transformation to improve the efficiency and effectiveness of their IT environments, cloud migration is often a high priority. Yet migrating workloads and applications can often be a challenging process. To help organizations efficiently make the move, here is a five-step cloud migration checklist gleaned from industry best practices. 1. Determine your […]