Latency from CPU_protocol for extremely bursty SAN workloads
Applies to
ONTAP 9
Issue
- High latency observed for extremely bursty SAN workloads. Breakdown of the latency shows high latency on Network & Protocol processing, this could be observed in various ways:
- Perfstat: Delay from
CPU_protocol
- PerfArchive: Delay from
Network
- Other performance monitoring tools like NetApp Harvest/Grafana will have similar latency breakdowns.
- Command Line:
qos statistics workload latency show
orqos statistics volume latency show
has the latency fromNetwork
- Perfstat: Delay from
Example:
- The high latency from
CPU_protocol/Network
is observed when there is no CPU contention in SAN processing CPU Domains, such asSSAN_Ex
.
Example: sysstat -M 1
(Diag Mode)
Note: If there is an overall CPU contention, then the delay from CPU_protocol/Network
is expected, the discussion here wouldn't apply.
- The following KB would help detect if there exists an extremely bursty workload
How to identify transient workload bursts using "sysstat" command