CPU is higher than expected on idle NVMe-oF AFF systems
Applies to
- ONTAP 9
- All AFF systems with NVMe drives in NVMe over Fabrics (NVMe-oF) configuration
- All AFF systems with NVMe drives in NVMe over TCP (NVMe/TCP) configuration
Issue
- CPU may idle around 10% or 20% when clients are connected but there are 0 client IOPs
- Example: sysstat -m at node shell from A400 shows cores 0-14 at 0%, but 15-19 at 40% (some rows removed for readability):
Cluster::> node run -node <node> sysstat -m 1
CPU15 CPU16 CPU17 CPU18 CPU19
44% 38% 39% 41% 38%
44% 37% 40% 42% 38%
45% 38% 39% 44% 38%
43% 37% 40% 41% 38%
44% 37% 39% 41% 38%
45% 45% 52% 46% 42%
43% 37% 39% 42% 38%
44% 38% 39% 40% 38%
44% 41% 40% 41% 38%
44% 38% 38% 41% 38%