Problem/Justification
mountd threads and nfsd threads don’t always scale together in all environments. For instance, in our environment we have many clients. with nfsd threads set to 128, the nfs daemon complains about too many connections. However, when we doubled the threads to 256 it also doubled the mountd threads to 64 which substantially increased the load on the TrueNAS system causing performance problems.
Impact
The default can still be to calculate mountd threads based on nfsd threads, but there should be an option to scale them independently.
User Story
I envision a user being able to scale their nfsd threads to service client I/O without scaling mountd threads when they don’t need it. This is especially pertinent on TrueNAS appliances which are limited in the number of CPUs that are available.