IBM Power Systems – The Road to Computational Longevity

By Charles King, Pund-IT, Inc.  November 7, 2018

Longevity is a trait seldom celebrated and less often achieved in IT. The former point is hardly surprising given the tech industry’s obsession with shiny new things. But the latter highlights how seldom even good solutions keep step with evolving technical trends and requirements, then lose the confidence of enterprise customers.

So how do long-lived technologies get and stay that way? By being innovative, dependable and adaptable, as the continuing success of IBM’s Power Systems attests. Since their introduction over two decades ago, Power Systems have been notable for their dynamic flexibility, resiliency and performance. Those characteristics made IBM Power solutions ideal for data-intensive workloads, including enterprise databases and associated applications.

They are also making IBM Power the platform of choice for newer and emerging use cases, such as in-memory databases, private cloud infrastructures and artificial intelligence (AI) deployments. In fact, those three areas helped Power Systems achieve 17% year over year (YoY) sales growth in IBM’s most recent (Q3 2018) quarter. How has IBM Power gotten from there to here? That’s a subject worth further discussion.

Focusing on data

Data has always been essential to businesses and the computational solutions they employ. That was true in the past, but today business information is key to successful competitive differentiation. At the same time, organizations are continually challenged by the “3 V’s” of modern data – unprecedented volume, variety and velocity.

How can enterprise vendors help clients overcome those issues? IBM has made certain that its Power Systems solutions are “built for data.” That is, they are designed to help IBM customers successfully address modern data challenges, deploy new applications and workloads, and harvest the full value of their business information assets. In addition, Power System solutions can support numerous other business and technical initiatives, including IT modernization efforts and universal encryption.

In fact, IBM Power Systems define the leading edge of enterprise performance in three key ways:

  1. Flexible – Power Systems are available in a wide range of scale-up and scale-out configurations. By design, these solutions can support numerous data-intensive scenarios, including mainstream applications, leading-edge HPC workloads and evolving artificial intelligence (AI) projects. Along with satisfying demanding business customers worldwide, IBM’s latest new POWER9 solutions are also the basis of the world’s first and third fastest supercomputers; the U.S. Department of Energy’s Summit and Sierra installations.
  2. Resilient – Power Systems have long been known for their RAS (resiliency, availability, serviceability) features, and IBM’s new POWER9-based solutions carry on that tradition. As a result, IBM Power is the platform of choice for some of the world’s most demanding telcos, banks and retailers. Power Systems are also designed to handle next-generation workloads and use cases, like blockchain.
  3. Fast – IBM POWER9-based systems can deliver up to 10X faster bandwidth acceleration and 50% greater memory bandwidth than comparable x86 solutions. Those capabilities help Power Systems speed training times for deep learning frameworks by up to 4X over x86 systems. Power Systems customers can also mix/match accelerator technologies (GPUs, FPGAs) with specific tasks and applications. Finally, IBM Power solutions support the latest in data transfer technologies, including PCIe 4.0 and novel NVLink and OpenCAPI interfaces, further enhancing their value for demanding applications and scenarios.

Along with these impressive characteristics, IBM Power Systems offerings are comparable to x86-based servers in price. However, their performance also typically exceeds x86 solutions, especially for applications that exploit large memory spaces, like in-memory databases, and those that benefit from robust I/O between the processor and system (including GPUs and other accelerators), such as AI workloads. Let’s consider these points in greater detail.

IBM Power meets new workloads

During the two decades since IBM introduced its POWER microprocessors and solutions, competing architectures and systems, including Oracle’s SPARC (inherited from Sun) and HP’s PA-RISC, have fallen into a hole or entirely off the map. Why is that the case? In large part because those vendors and platforms failed to keep up with the changing needs of their customers.

In sharp contrast, IBM has continually and actively evolved Power Systems to maximally support both existing business applications and new workloads and use cases. Today, those efforts include:

  • SAP HANA in-memory databases – IBM’s initial Power for HANA solutions were introduced in 2015 and the company currently boasts 1600+ Power for HANA customers. IBM’s offerings are unique alternatives to conventional HANA solutions—all of which are based on x86 hardware. IBM Power Systems’ capacious memory, massive bandwidth, overall performance, and ability to quickly ingest/analyze large data sets make them excellent, cost-competitive platforms for HANA implementations. Additionally, since IBM and SAP are longstanding strategic partners, the companies’ common customers can easily obtain the assistance they require for HANA on Power deployment and support.
  • Private clouds – As cloud computing matures, enterprises are increasingly utilizing multiple cloud platforms. Power Systems servers, solutions and services can be quickly and effectively deployed in private cloud, hybrid cloud, multi-cloud and public cloud scenarios. Since Power Systems are designed from the ground up to secure organizations’ critical applications and assets, and to simplify the movement of data, applications and services, it should come as no surprise that the vast majority of top ten banking, retail, insurance and telco companies all depend on IBM Power.
  • AI – Machine Learning/Deep Learning – IBM’s POWER9 is designed to handle extreme data-intensive applications, from enterprise transaction processing to demanding machine learning and deep learning processes central to AI, in the world’s fastest supercomputers – the DOE’s “Summit” and “Sierra” installations. In addition, IBM created PowerAI – enterprise-class distributions of AI applications (TensorFlow, Chainer, Café, etc.) to enabling data scientists to quickly turn data into valuable insights. IBM’s partnership with NVIDIA resulted in memory-coherent NVLink solutions which, along with IBM’s own OpenCAPI, can accelerate performance bandwidth up to 10X faster than the PCIe 3.0 technologies that x86 solutions typically support. The tools IBM and its partners have created for Power solutions enable customers to deploy AI models that are significantly larger and more complex than many other AI platforms.

Additionally, in 2014, IBM launched “Linux on Power” – an initiative designed to bring the full value of the company’s Linux and open source investments to Power Systems. Linux on Power includes open source development tools (compiler, debugger and profiling tools) and runtime libraries. IBM ISV partners and developers have optimized over 200 enterprise solutions for Linux on Power, with additional offerings in the pipeline.

Add-in another key benefit; that the 95% of Linux on x86 applications written in C/C++ can be ported to Linux on Power systems with no source code changes—just a simple recompile and test. That’s great news for enterprises, so it should come as no surprise that IBM reports its Linux on Power solutions sales are growing at 3X the rate of the total Linux server market.

IBM Power customers

How are IBM customers putting Power Systems to work in these new use cases and scenarios? Here are three examples:

  1. The Wurth Group – SAP HANA: The Wurth Group is a leading global retailer of assembly and fastening materials. Like many high volume/low margin businesses, it faces stiff competition from online shops. To become faster and more efficient, the company decided to integrate its SAP ERP and database environment (running on IBM Power Systems) and its SAP HANA analytics system (running on an x86 cluster) by implementing an IBM HANA on Power solution. Doing so resulted in numerous benefits: the number of physical systems required for SAP HANA was reduced by 86% (one Power Systems server replacing seven clustered x86 servers); the time required to upgrade the full stack was reduced by 43% – from a full week to a few days; Power Systems’ advanced virtualization enabled IT to change the size of SAP HANA databases dynamically by adding compute and memory as needed. Along with running SAP HANA alongside its other SAP applications, the IBM Power solution also enables the company to provide updated analytics and reporting daily—critical to finding new clients and up-sale opportunities. With IBM’s help, Wurth Group has fully empowered its sales organization.
  2. CipherHealth – Private Cloud: CipherHealth is a software provider whose solutions are designed to help healthcare professionals from clinicians to administrators work together more efficiently, improve patient outcomes and cut costs. Like many growing organizations, CipherHealth initially took advantage of hosted services to scale its solutions and improve redundancies. However, the company found that the oft claimed benefits of public cloud – reduced costs, ease of use, run-time scaling – never fully materialized. The company structured real world tests to see whether an on-premises solution would work and found the metrics were “off the charts” in terms of speed and reduced runtime variability. Working with IBM, CipherHealth moved forward with an implementation leveraging Power Systems and Linux on Power solutions. The new infrastructure is providing greater transparency and performance. For example, BI (business intelligence) extract, transform and load time processes have dropped from six hours to 45 minutes. Consistent performance has also reduced pressure on CipherHealth’s IT team, helping them to focus on value-added tasks. The company also plans to leverage Power solutions to support continuous integration testing. Overall, CipherHealth reports that IBM’s Power Systems are delivering five times the performance and scale of its prior cloud-hosted infrastructure at half the cost.
  3. Wells Fargo – AI: While artificial intelligence (AI) initiatives often show great promise, many hit obstacles as they move from POC to delivering optimized and trained models quickly enough to benefit the organization. Achieving the final stage – integration across multiple business lines – is even harder since it requires an enterprise-scale infrastructure that offers granular integration, security and support features. It isn’t enough to offer models one at a time; business units want multiple results immediately, even as data volumes are rapidly growing in size and complexity. One enterprise that is successfully navigating these AI challenges is Wells Fargo, which, with IBM’s help is using deep learning models to comply with critical financial validation processes. Using IBM Power Systems, IBM PowerAI Enterprise software, IBM PowerAI and IBM Spectrum Conductor (the company’s next generation software-defined infrastructure solution optimized for Apache Spark), Wells Fargo’s data scientists are building, enhancing and validating hundreds of models daily. As Richard Liu, Wells Fargo’s Quantitative Analytics manager, said at IBM Think 2018, “People talk about fancy (AI) algorithms but in real life, how efficiently the models run in distributed environments is critical. IBM is a very good partner and we are very pleased with their solution.”

Final analysis

Long-lived IT solutions get and stay that way by being innovative, dependable and adaptable. In doing so, they retain the trust of existing customers and provide attractive alternatives to prospective clients. IBM’s Power Systems are a good example of this dynamic. By leveraging the company’s deep experience with enterprise clients and its ongoing hardware, software and silicon investments, Power Systems solutions have remained at the innovative edge of enterprise computing.

IBM’s newest Power Systems solutions featuring its POWER9 silicon are helping continue that success. A careful study of the newest Power Systems offerings demonstrates that the company’s “built for data” appellation isn’t mere marketing. Along with optimally supporting traditional enterprise workloads, these latest solutions’ inherent flexibility, resiliency and performance make them ideal platforms for data-intensive use cases, including SAP HANA, private and multi-cloud deployments and AI-related processes.

IBM customers, including The Wurth Group, CipherHealth and Wells Fargo fully understand the tangible, measurable value that Power Systems delivers, and are actively garnering those benefits. Enterprises not already working with IBM would do well to investigate Power Systems solutions as they consider their current and future IT investments.

© 2018 Pund-IT, Inc. All rights reserved.