Sequential workloads

Measurements were collected when running sequential read and sequential write workloads while scaling the number of jobs configured for each run. The workload was run directly on the KVM guest. These results were observed:

• Low workload job counts with sequential read were affected by the KVM guest page cache, with high cache hit ratios providing high throughput numbers with a low amount of real disk I/O, which lets the guest stay in the SIE execution state. There are minor differences between the attachment modes; NFS has a slight advantage in regard to transferred throughput per CPU.
• For each attachment type, sequential write shows different strengths and weaknesses regarding throughput. SMB performs well at the low workload level; iSCSI performs well for a medium workload level; NFS shows clear strength in the high workload level area.
• For throughput transferred per CPU (CPU efficiency), each of the protocols were closely clustered, with NFS exhibiting small leading values.