Ask someone what they expect from their virtualized environments and they’ll probably tell you that they want to save money while meeting service level objectives. This is done by utilizing available platform resources to the max while ensuring that high-priority workloads are being processed ahead of lower-priority ones.
I just read this great article and wanted to briefly call out a specific set of data points because they encapsulate why companies should be considering System z for consolidating Linux workloads.
IBM ran a test to compare the effect of adding additional workloads to a running virtual environment on System z vs. Intel. They initiated a set of high-priority workloads on both platforms and measured the CPU utilization, throughput levels and response times. Then they added some low-priority workloads and measured the effect on the running, high-priority workloads.
Note that the additional workloads were low-priority; the desired result was that these additional workloads should be processed only when resources were available and that the high-priority workloads should not be adversely affected. System z delivered on this objective beautifully but the Intel systems failed to deliver across the board.
The bottom line is that the System z “demonstrated nearly perfect workload management” but the Intel based solution struggled to manage the low-priority tasks without affecting the high-priority tasks. The conclusion is that on the Intel platform you have to either over-provision resources (costs $$s), risk not meeting your service level agreements (costs $$s and reputation)…or just use System z!
Comments