We all remember that scene from the Wizard of Oz where Toto pulls the curtain back to reveal the Wizard of Oz pulling levers and spinning wheels. He roars into his microphone – “Pay no attention to that man behind the curtain!”
This scene from the famous 1939 movie starring Judy Garland provides a fun analogy from popular culture for the various IT technologies now collectively known as – virtualization.
Even though the actor Frank Morgan playing the Wizard of Oz character turns out to be less magical than Dorothy hopes, in real data center life, virtualization has demonstrated extraordinary powers.
High performance computing (HPC) is one area that has leveraged the benefits of virtualization for decades. From early in the days of large mainframe processors, engineers created components whose task, among many others, was to pretend to be something they were not, in order to solve certain IT challenges or provide value to end users in some way or another. For example, as early as 1976, IBM was experimenting with logical partitioning (LPAR) – dividing a computer’s processors, memory, and storage into multiple sets of resources so that each could be operated independently, with its own operating system instance and applications. Logical partitioning in various forms continues to be used today in everything from mainframes to hard disk drives.
The concept is simple – separate or “abstract” the actual physical resource from the rest of the solution components. Most often, this is accomplished by adding a layer of software between the end user, such as an application, and the IT resource. The application still believes it’s running on a certain processor and requesting data from a certain storage array – but it’s not. The software or virtualization layer is pretending to be those physical entities – for some very good reasons.
From HPC and big data analytics to today’s Artificial Intelligence (AI), these workloads have moved to compute infrastructures composed of multiple systems, sometimes several, sometimes dozens, sometimes hundreds or even thousands. And these complex infrastructures often run very large compute workloads that are broken down into many smaller tasks, spread across many compute nodes, and run in parallel to more rapidly and efficiently solve the world’s most intractable problems. Managing all this complexity, organizing it to properly support each application and workflow, and then sorting things out when one of the systems or compute nodes fails is the stuff of many a systems administrator’s nightmares…
…unless you virtualize the entire collection of systems or “cluster” so that to the applications and the human end users it all appears like simply one big – resource.
Through the wizardry of cluster virtualization, a compute node can fail and the show will go on without a hitch. With a wave of the virtualization wand, work will run on the right node, with the right resources, at the right time. A top priority task might be given permission to hog most of the cluster resources, and all the other tasks will patiently wait their turns. A number of resource hogs may all be scheduled to run at the same time, to the point where resources run low, and a really great cluster virtualization wizard will marshal additional resources from that place of rainbows – the cloud. While you, as the end user or administrator, don’t need to be a wizard at all to make it happen!
And this highlights a very important fact – not all cluster virtualization solutions are equally powerful. IBM Spectrum Computing has been running some of the world’s most demanding workloads on shared compute clusters for over two decades. The family of software-defined infrastructure solutions is widely viewed as the management software of choice for technical and high performance computing in computationally and data-intensive applications such as real-time analytics, AI, deep learning and machine learning, simulations, product design analysis, and risk management in industries and public settings as diverse as financial services, manufacturing, digital media, oil and gas, life sciences, government, research, and education.
IBM Spectrum Computing products optimize distributed computing environments to deliver faster results while making them easier to manage and scale. They include functionality and capabilities to ensure that jobs and data sets fail over without complications when individual cluster nodes go down or are taken down. The family includes powerful resource schedulers that enable you to tame the processing hogs while maintaining high resource utilization. And very exciting in this 21st century world of rapidly evolving requirements, IBM Spectrum Computing provides hybrid cloud bursting capabilities so that your HPC infrastructure can grow and shrink intelligently, based upon workload demand, to minimize how much cloud you use in order to control costs.
IBM Spectrum Computing brings plenty of magic to cluster virtualization. When you set off on your next big data processing, analytics, or deep learning adventure, remember that IBM technologies and solutions can help get you home with only a few heel – or mouse – clicks. And you won’t need to capture the broom from the Wicked Witch to pay for it.
Here’s a quick video to help get you started.