Economics and Performance Make Linux on System z the Clear Choice
Data centers run a range of business workloads, including batch and transaction processing, business applications, complex data analysis, collaboration and social business. It’s easy to gravitate toward one particular server as being good for all of these workloads; however, they all have different requirements. For that reason, IBM offers different types of servers.
It’s essential to understand what hardware is best suited for which applications and why. Determining the best placement of business workloads should focus on the platform that delivers the best performance at the best cost with the best quality—a concept known as “best fit.” Consider these three workload classes and see why the best fit is on IBM System z*:
1. Linux Consolidations
Several characteristics make the mainframe a highly efficient platform, especially for consolidating Linux* workloads. Its core features—processing power, large cache and a dedicated I/O subsystem—provide superior scalability and throughput.
In addition, exceptional workload management, when combined with the aforementioned capabilities, makes the mainframe especially good for consolidating Linux workloads. Workload management refers to the effectiveness and efficiency of the virtualization layer to manage resources across multiple workloads.
To realize the best cost per workload, a platform must maximize CPU utilization and ensure high-priority workloads always meet their service levels, particularly when a mix of high- and low-priority workloads are running simultaneously.
To demonstrate the superior workload-management capability of System z, IBM ran tests comparing mainframe virtualization with a common distributed server hypervisor running on an Intel* technology-based server. Stand-alone high-priority workloads were run on both platforms to measure their CPU utilization, throughput levels and response times. Then the high- and low-priority workloads were run concurrently to see how the virtualization layers managed resources.
On System z, the high-priority workloads maintained their CPU utilization, throughput levels and response times, with the low-priority work consuming all but 2 percent of remaining CPU minutes. On the Intel technology-based server, utilization of high-priority workloads dropped 28 percent, throughput dropped 31 percent, and average response times increased 45 percent when the low-priority workloads were added. It also had 22 percent unused CPU minutes.
These tests demonstrated nearly perfect workload management on System z. Meanwhile, the Intel server and hypervisor’s imperfect workload management was unable to handle the workloads effectively without adversely affecting the high-priority ones. Accordingly, the Intel technology-based platform requires segregating workloads onto separate servers to correctly manage them, leading to core proliferation and increased costs.