IBM takes mainframe to the forefront of cloud with IBM z16 and LinuxOne rack mount configurations

Omdia view

Summary

In May 2023, IBM released its latest mainframe products, the IBM z16 and LinuxONE Rockhopper 4 single frame and rack mount models. These new configurations allow the IBM z16 and LinuxONE to coexist with other servers in a standard 19-inch rack. This also marks the first time that IBM is shipping mainframe systems without an integrated rack. IBM trimmed the mainframe formfactor from a 24-inch chassis to 19-inch rack back in 2017 with the launch of z14 systems and, now, leaves the rack choice to customers. This move is expected to simplify the deployment of mainframes in data centers and colocation facilities, allowing them to integrate seamlessly with other computing infrastructure. IBM has integrated Red Hat OpenShift with IBM z16 and LinuxOne, enabling cloud native software development and deployment on mainframes. Availability of these new hardware platforms at lower price points, flexibility in configurations, and a cloud native software platform would attract a wider range of developers to leverage the hardware capabilities and create new opportunities for IBM in the market.

The enduring relevance of mainframes in today's technology landscape

Despite the increasing popularity of cloud computing, mainframes still play a crucial role in many organizations' IT infrastructure. According to IBM, IBM Z and LinuxOne run about 70% of global financial transactions by value. Over the last decade, the installed capacity of IBM mainframes, measured in million instructions per second (MIPS), grew over three times. IBM’s customers are upgrading their systems too. According to IBM, two thirds of the z servers and 85% of the installed capacity (in MIPS) is on z14 or newer technology; meaning that the servers are less than five years old. The reliability, security, and scalability offered by mainframes makes them the preferred compute platform for mission-critical applications. IBM continues to invest in the development of new mainframe models with more features and capabilities to meet the changing needs of the market. Here are some key features of IBM Z and LinuxOne that make it an efficient and reliable host for database applications and transaction processing:

• High-speed transaction processing: IBM Z and LinuxOne are designed to deliver high-speed processing capabilities with their new Telum processors and memory subsystems. The CPU cores are an out-of-order design with simultaneous multithreading and can operate at or above a 5GHz base frequency. Additionally, embedded features like zEnterprise Data Compression (zEDC) and IBM Z Sort helps to increase data transfer rates to boost throughput without adversely impacting response times of workloads like online transaction processing (OLTP).
• Large memory capacity: Both these server models have a large memory capacity, up to 40 TB of Redundant Array of Independent Memory (RAIM) per system, allowing it to handle massive amounts of data. The IBM Telum CPU also has semi-private 32 MB level-2 (L2) caches per core that are working in concert to provide up to 256 MB virtual Level-3 (L3) cache per chip and up to 2 GB virtual level-4 (L4) cache per server. This is particularly important for database applications that require fast access to large amounts of data and there aren’t many server options available in the market with such large memory capacity and redundancy.
• Integrated security: The Telum CPU features advanced security capabilities that are built into the hardware. These include encryption, access controls, and secure key management, ensuring that sensitive data is protected from unauthorized access or tampering. Quantum-safe cryptography is embedded in the system to improve the resiliency to cyber-attacks. IBM Z Cyber Vault extends the cyber resiliency of the IBM z16 by enabling the quick recovery from ransomware attacks.
• High availability and fault tolerance: IBM Z and LinuxOne are designed to provide high availability and reliability, with features such as redundant components, automatic failover, and disaster recovery capabilities. This ensures that database applications and transaction processing workloads are always available, even in the event of hardware or software failures. IBM z16 servers are designed for 99.99% availability, or 3.16 seconds of downtime per year, and again, there aren’t many other alternate server hardware options in the market offering such reliability levels.

The new single frame and rack mount configurations for zSystems and LinuxONE makes it easy to deploy mainframes in a distributed, hybrid cloud environment, allowing easy integration with storage and compute resources within the client's own racks. This opens doors to expanded use cases for mainframes and provides end-users with enhanced flexibility and control over infrastructure deployment.

The software stack on IBM Z and LinuxOne have also undergone serious transition. While legacy mainframe applications written in COBOL continue in use, developers today have a wide range of programming languages and tools available to choose from, depending on their specific needs and requirements. IBM Open Enterprise SDK for Python, enables modernizing mainframe applications by developing APIs, plugins, and wrappers. For AI/ML workloads, the IBM Z deep learning compiler (IBM zDLC) enables compiling deep learning models into shared object libraries that can be used on Linux, IBM Z and LinuxONE (see Figure 1). IBM zDLC is based on the open source open neural network exchange’s (ONNX) multi-level intermediate representation (MLIR) project and can be used to compile models from a variety of deep learning frameworks, including TensorFlow, PyTorch, and Keras.

Figure 1: IBM zDLC pipeline Source: IBM

Hybrid Cloud with IBM Mainframe

Ove the last few years, specifically after the COVID-19 pandemic, there is a big surge in enterprise digital transformation initiatives. It has become inevitable for enterprises to leverage digital technologies, strategies, and initiatives to fundamentally change their operations, processes, and business models for agility, cost optimization, and efficiency. The mainframe applications are also undergoing significant reengineering to enhance agility, optimize investments, and drive accelerated innovation, and this is where IBM’s acquisition of Red Hat played a critical role. IBM is leveraging Red Hat OpenShift and IBM Cloud Paks to provide developers with a platform for application modernization with pre-integrated data, automation, and security capabilities. The IBM Z digital integration hub enables real-time information flow and integration between mainframe databases and hybrid cloud applications. IBM Z digital integration hub is also integrated with leading cloud services providers such as AWS, Microsoft Azure, Google cloud platforms, and IBM Cloud, making the job of developers easier.

IBM zSystems and LinuxONE also integrates with popular DevOps tools and frameworks, such as Jenkins, Git, and Ansible. This integration allows for streamlined continuous integration and continuous deployment (CI/CD) pipelines, automated testing, and version control for mainframe applications.

Sustainability of Mainframes

Sustainability and energy efficiency is a critical concern for data centers. Apart from optimizing the datacenter physical infrastructure for optimal use of energy and other resources, such space and water, operators are also looking into ways of doing computing efficiently. Different workloads have varying levels of computational intensity and power requirements. Optimizing datacenter workloads for energy efficiency also involves using a heterogeneous mix of processors that strike the right balance between performance and energy efficiency for specific workloads. Specialized processors like GPUs, TPUs, FPGAs, and other ASICS often offer higher performance per watt compared to general-purpose CPUs for specific workloads.

Mainframes have long been recognized as a preferred platform for transaction processing, especially for highly secure and mission critical workloads. IBM has optimized the Telum CPU used in z16 and LinuxOne systems for every aspect of transaction processing. It has a bigger and faster cache memory offering improved latency and, along with optimized cores, Telum provides over 40% improvement in performance over its previous generation of CPUs according to IBM. The CPU has embedded accelerators for sort, compression, and cryptography along with a dedicated accelerator for AI inference primarily targeting use cases like fraud detection, customer behavior prediction etc.

IBM claims, when running WebSphere and DB2 workloads, LinuxONE Emperor4 requires 16 times fewer cores than the compared x86 servers, and the full configuration LinuxONE Emperor4 Max125 would be doing the work of about 2000 cores of the compared x86 servers. According to IBM this will result in a reduction of energy consumption by 75%, and datacenter floor space by 50%.

Considering an average of 24 cores on a general purpose x86 socket, that is equivalent to twenty 4-socket x86 servers, costing a total of over $700 thousand. IBM has not provided the details of the exact server configuration used and has extrapolated the per core performance benchmarks to rack scale for the comparisons and it may not be drawing an accurate picture. However, undoubtedly z-16 and LinuxOne are systems optimized for transaction workloads and could provide significant performance advantage over standard x86 CPU bases systems. While deploying critical workloads on x86 systems, it would become inevitable to have redundant servers to ensure reliability which is not the case with IBM’s Mainframes. There could be significant reduction in operational expenses as well with z-16 and LinuxOne because the redundant x-86 servers could end up using more power and floor space. Given these factors, z-16 and LinuxOne may be a good choice for transaction-focused workloads and may offer efficiencies that align with customers' sustainability goals.

Bottom line

Datacenter computing is in transition, and the drive for efficiency and sustainability is making it more heterogeneous than ever before. There is significant opportunity for highly reliable and efficient compute systems in datacenters and IBM z16 and LinuxONE4 are proven hardware platforms when it comes to reliability and efficiency for transaction processing workloads. The integration of different Red Hat technologies such as RHEL, OpenShift, and Ansible has made the software development environment for these mainframe servers more like that of x86 systems, eliminating the developer friction towards mainframes. The new rackmount configurations are providing more formfactor choices to end users and makes deployment easier, along with their existing infrastructure on-premises or with co-location vendors. Omdia believe all these could attract a more diverse set of end users to consider IBM mainframes as a platform for their transaction workloads and could offer a significant tailwind for IBMs server revenues.

Appendix

Author

Manoj Sukumaran, Principal Analyst, Datacenter, Compute, and Networking

askananalyst@omdia.com