
Key facts at a glance
- IBM is positioning IBM Z and LinuxONE as native participants in Ethernet and IP networks rather than isolated mainframe networking islands.
- The strategy combines conventional Ethernet and IP support, more remote direct memory access options, higher bandwidth, lower latency, and tighter integration with cloud, storage, distributed systems, and AI infrastructure.
- IBM has moved I/O acceleration onto an on-chip Data Processing Unit integrated with the Telum II processor, reducing reliance on custom ASICs and freeing main processor cores.
- A 100GbE adapter for IBM Z and LinuxONE adds two 100GbE ports with QSFP SR optics and plugs directly into the central processor complex drawer.
- The Network Express converged adapter combines OSA Ethernet, RoCE and RDMA, and Coupling Express 3 Long Reach functions on one physical adapter.
- Enhanced Queued Direct I/O redesigns the Ethernet data path for the z17 generation and its on-chip I/O processing.
- Security additions include automated network micro-segmentation and AI-driven anomaly detection for z/OS environments.
- New management tools target I/O topology visualization and infrastructure as code using Terraform and Ansible-friendly workflows.
From specialized networking island to hybrid infrastructure participant
IBM Z has traveled a long way from its roots as a specialized networking island. For years, mainframe connectivity depended heavily on dedicated channel technologies, specialized adapters, and operational practices that were distinct from the Ethernet and IP world used by distributed systems. That separation had benefits in reliability and control, but it also made the mainframe look different from the rest of the data center. The modern IBM Z design instead treats the mainframe as a high-performance participant in a broader enterprise network, using standard Ethernet and IP while adding technologies for low-latency and high-throughput communication.
The current networking strategy is built around three practical goals. First, IBM wants strong support for conventional Ethernet and IP connectivity so the mainframe can fit into ordinary enterprise network operations. Second, it wants to increase remote direct memory access based networking options for workloads that need faster movement of data with less overhead. Third, it wants secure, proficient communication with the large data center environment that often surrounds the mainframe, including cloud platforms, distributed servers, storage systems, and AI infrastructure. IBM executives describe the aim as making IBM Z a deeply integrated part of a client hybrid infrastructure, with continued improvements in bandwidth, latency, and I/O capacity.
That shift matters because mainframe workloads no longer sit alone. Transaction processing, batch operations, Linux workloads, analytics, and AI pipelines increasingly depend on data and services that live on other platforms. A mainframe that can connect cleanly to Ethernet fabrics, storage networks, and cloud gateways is easier to include in modern resilience, automation, and security designs. The networking roadmap is therefore not a side project. It is part of keeping the platform relevant as enterprise architecture changes around it.
On-chip DPU offloads I/O work
One of the clearest examples of the mainframe networking evolution is the decision to offload I/O processing onto purpose-built technology. IBM has integrated a Data Processing Unit architecture with the Telum II processor design that underpins current IBM Z systems. The DPU includes four processing clusters, eight programmable microcontrollers per cluster, an I/O accelerator that coordinates the clusters and manages I/O subsystems, and a dedicated request manager. That design is intended to handle large I/O workloads without forcing the main processor cores to spend cycles on routine I/O overhead.
Offloading I/O is not a new idea on the mainframe. IBM has used custom ASICs on I/O cards for years to keep the main processor focused on computation. The move to an on-chip DPU changes the economics and the cadence. Clients still avoid spending Z core cycles on I/O functions, but the on-chip design is expected to deliver better performance, latency, and throughput. It also removes many of the older ASICs from the system. Depending on card types and client requirements, those ASICs could number in the hundreds, so removing them can save power and simplify the physical design.
There is a density benefit as well. With fewer ASICs on the I/O cards, IBM can move to high-port-density cards for protocols such as Fibre Channel and FICON. Fewer cards can mean fewer I/O drawers, and in some configurations it can reduce the number of Z racks or frames. The engineering model also changes. Older I/O ASICs were released roughly once a decade and required large additional investment, which created technical debt between generations. By pulling that functionality into the main processor chip, which is refreshed with every mainframe generation, IBM gains a chance to innovate more frequently and deliver new technology to clients faster.
100GbE targets Linux consolidation and hybrid cloud access
Another recent upgrade is a 100GbE adapter for IBM Z and IBM LinuxONE systems. The adapter features two 100GbE ports with QSFP SR optics and plugs directly into the central processor complex drawer, eliminating the need for external I/O drawers for high-speed network connectivity. That direct connection is important for clients that want to consolidate large numbers of Linux virtual machines onto a mainframe while keeping network throughput high and the physical footprint under control.
LinuxONE is particularly associated with this consolidation story. IBM urges customers to move hundreds or thousands of Linux virtual machines onto a single mainframe, turning a sprawling collection of distributed servers into a centralized network server. The mainframe is I/O heavy compared with a typical 1U server, and that is a deliberate design choice. To handle critical workloads with high resiliency, the platform must move a lot of data in and out at speed, with redundancy at every layer. A system with that much redundancy is less bandwidth constrained than a small server with one or two network interface cards.
The 100GbE request came primarily from clients running Linux on the platform, including LinuxONE and Linux on Z customers. The platform offers sustainability benefits by enabling large workload consolidations, for example reducing thousands of x86 cores to a few hundred Linux cores. IBM has cited a global bank that migrated sprawling x86 installations to a few LinuxONE servers and gained better performance, security, and sustainability savings. Other clients want to repeat that pattern with more Linux workloads, and some said the lack of 100GbE prevented them from connecting into the parts of their hybrid cloud environment needed to compete for those workloads. The new adapter removes that blocker.
Converged adapters and EQDIO simplify the network design
IBM also introduced the Network Express converged adapter with the z17 family rollout in 2025. The adapter melds two previously separate networking architectures, OSA Ethernet and RDMA over converged Ethernet, and adds support for the Coupling Express 3 Long Reach function on the same physical adapter. It supports transmission speeds from 10G to 25G and can link with single-mode and multi-mode fiber. The coupling function includes a two-port Ethernet-based long-distance coupling card for tying together remote operations.
Converged adapters enable further physical I/O consolidation and give customers more flexibility with less hardware. That can reduce the hardware configuration burden and potentially lower the number of network connections to manage. Network Express also uses Enhanced Queued Direct I/O to speed Ethernet traffic. The older Queued Direct I/O technology was designed for the Ethernet speeds and I/O architecture of earlier Z generations. Enhanced Queued Direct I/O redesigns that path to take advantage of the z17 on-chip I/O processing and to handle higher-throughput, lower-latency Ethernet traffic more efficiently.
Taken together, these changes move the mainframe away from multiple specialized adapters and toward a converged Ethernet-based I/O architecture. In that architecture, conventional IP networking and high-performance RDMA share the same physical infrastructure. The direction is consistent with broader data center trends, where converged and composable infrastructure reduces cabling complexity and gives operations teams fewer moving parts to manage.
Security and segmentation become more urgent
As the mainframe becomes just another part of the modern networking environment, security demands increase. Industry analysts note that the mainframe is no longer strange or unusual in network terms, which means it is exposed to the same operational and security pressures as other critical systems. Segmenting and securing the network around the mainframe needs more attention than it once received. Annual penetration tests may have been acceptable in the past, but that cadence is not strategic for a platform tied to high-value transactional data. Patching also needs to keep pace with a faster threat environment.
IBM has responded with the zSecure Detection package, which combines threat monitoring, network insights, AI-driven access anomaly detection, and automated response for z/OS. The package includes automated network micro-segmentation capabilities that generate customized segmentation policies based on observed networking behavior. Those policies can help organizations reduce unnecessary connectivity, limit opportunities for lateral movement, and strengthen security controls across critical environments. The feature fits the broader networking theme because segmentation depends on accurate visibility into how systems, adapters, fabrics, switches, and endpoints connect.
AI keeps the mainframe close to transaction data
IBM sees a continuing role for the mainframe as AI spreads through enterprise technology. Enterprise AI will rarely operate on one platform alone. Models, applications, and data will be distributed across mainframes, cloud platforms, accelerators, and other infrastructure. The mainframe differentiated role is its ability to apply AI close to highly sensitive, high-volume transactional data that already resides on the platform. That proximity can reduce unnecessary data movement while supporting faster decisions and stronger data governance.
For IBM, the mainframe remains a significant AI platform through a combination of on-platform AI capabilities, high-speed networking, and interoperability with the wider AI ecosystem. Better networking is not just about raw speed. It is also about making it practical to move training data, model inputs, results, and operational telemetry between the mainframe and other systems without creating new bottlenecks or governance gaps. The 100GbE adapter, converged adapters, and DPU design all contribute to that practical integration.
Visualization and infrastructure as code extend the modernization story
Beyond networking, IBM has introduced tools aimed at helping customers manage and grow mainframe infrastructure. The IO Topology Visualization tool creates a graphical map of how an IBM Z system I/O infrastructure is physically and logically connected. It shows connections among systems, adapters, fabrics, switches, and endpoints. The tool is part of the Infrastructure Management for Z and LinuxONE package. It addresses a long-standing pain point: clients have often built detailed diagrams by hand in presentation or diagramming software, which is manual, painful, time-consuming for experienced team members, and error prone. Those diagrams are still used for planning new devices, troubleshooting, and showing resiliency status to leadership, so they must be created and kept current. Automating the first release of that visualization reduces the burden and creates a foundation for more features.
The second new technology is Infrastructure as Code via IBM Infrastructure Manager and IBM Terraform. Configuring mainframe infrastructure gives users granular control, which is valuable, but specialized tooling requires specialized skills. The Infrastructure Management offering for IBM Z and LinuxONE recently became generally available, providing a modern interface that pairs with Terraform to let users apply Infrastructure as Code on the platform. Infrastructure as Code is common in distributed computing, and younger professionals are entering the workforce with those skills, which can shorten time to proficiency compared with learning bespoke tools from scratch. The approach allows configuration of I/O resources and logical partitions while working with industry automation such as Ansible playbooks. That can make environments repeatable, so identical setups can be created in hours rather than weeks or months.
Source:Network World News
