Heavy SMB signing pressure causes high SMB3 latency from Network CPU domain

ONTAP 9

SMB3

SMB2

High SMB latency from Unified Manager, Grafana, Perfstat/PerfArchive. In the latency breakdown section, the bulk of the latency is from CPU_network or CPU N-blade, depending on the tools that are used to monitor performance
The impacted Vservers have a big amount of SMB3 or SMB3.1 connections
- cifs:vserver_name:total_smb3_connections_count:100000
- cifs:vserver_name:total_smb3_1_connections_count:100000
The impacted Vservers have signed sessions, as well as a significant amount of data being signed per second
- cifs:vserver_name:signed_sessions:50
- smb2:vserver:bytes_signed_rate:20000000b/s
Network CPU is not necessarily saturated
- Avg CPU util might NOT be high
- Nwk_Exmpt CPU domain util might not be higg
Other protocols might be impacted as well, such as NFS or iSCSI. As they have to complete for the same network threads with SMB signing workload
The util of NwkThd threads isn't necessarily high in ps
The average on queue time for NwkThd threads isn't necessarily high in ps -s

   ID   RunCount   RunMax  RunAve   OnQMax OnQAve Pri D Name
  272    2469043     2498      21    39726      3 100 N NwkThd_00