NFS3 Client Commands Interval Statistic

This feature collects statistics on the host and application side by sending NFS requests sequentially, introducing a delay between each request, and aggregating the results based on source IP and destination IP pairs according to the delay duration.

example:

One nfs request (rpc.procedure 1 is GetAttr) was delayed over 1000ms to send during communication between client 10.174.242.225 and nfs server 10.231.192.50

Example trace

Client sent ack to NFS server from TCP layer instantly then delayed 5 seconds before sent another NFS request.

How to filter in wireshark trace:

ip.addr == 10.174.242.225 && ip.addr == 10.231.192.50 && tcp.time_delta > 1000 && rpc.msgtyp == 0

Workaround:

It usually indicates bottleneck on host/application side.

sunrpc.tcp_max_slot_table_entries in Linux:

sunrpc.tcp_max_slot_table_entries is a Linux kernel tunable that controls the maximum number of concurrent (in-flight) RPC requests per TCP connection for SunRPC-based protocols, most notably NFS over TCP.

What it does

It defines the upper limit of the RPC slot table size for a TCP connection.

Each “slot” represents one outstanding RPC request that has been sent but not yet replied to.

Applies to NFS client ↔ NFS server communication over TCP.

Why it matters

A higher value allows more parallel NFS requests, which can:

Increase throughput on high-latency or high-bandwidth links

Improve performance for workloads with many outstanding I/O operations

A lower value:

Limits concurrency

Can reduce memory usage and protect servers from overload

Typical behavior

Default value depends on the kernel version (commonly 64 or similar).

The effective number of outstanding requests is also constrained by:

NFS mount options (e.g., nconnect, rsize, wsize)

Server-side limits

TCP windowing and congestion control

Example

If:

sunrpc.tcp_max_slot_table_entries = 128

Then:

Up to 128 concurrent RPC calls can be outstanding on a single TCP connection to an NFS server.

How to view or change it

# View current value

cat /proc/sys/sunrpc/tcp_max_slot_table_entries

# Temporarily change it

sysctl -w sunrpc.tcp_max_slot_table_entries=128

# Make it persistent

echo "sunrpc.tcp_max_slot_table_entries=128" >> /etc/sysctl.conf

Key Windows knobs that affect NFS concurrency

NFS Client thread limits

Windows NFS client uses worker threads to issue RPCs.

Registry path:

HKLM\SOFTWARE\Microsoft\ClientForNFS\CurrentVersion\Default

Common values:

Threads – number of NFS client worker threads

MaxThreads – upper limit for NFS threads

Effect:

More threads ⇒ more concurrent NFS RPCs

Too many threads ⇒ CPU contention

This is the closest functional equivalent to increasing RPC slots on Linux.