Rock 5 ITX optimizations : Samba and Dual 2.5Gbe port.

99-nas-network.conf

# /etc/sysctl.d/99-nas-network.conf
net.core.rmem_max           = 33554432
net.core.wmem_max           = 33554432
net.ipv4.tcp_rmem           = 4096 87380 33554432
net.ipv4.tcp_wmem           = 4096 87380 33554432
net.ipv4.tcp_window_scaling = 1
#vm.swappiness=10
#vm.vfs_cache_pressure=50

routing-issues.md

Fixing Asymmetric Routing & TCP Throughput Issues with Policy-Based Routing

Problem

A server with two interfaces on the same subnet was binding iperf3 to each address, but all return traffic used the lower-metric interface. This caused asymmetric routing and capped each flow at ~470 Mbps, even though each NIC could do ~940+ Mbps independently.

• Interfaces & Metrics
  • enP3p49s0 → 10.0.7.6 (metric 100)
  • enP4p65s0 → 10.0.7.5 (metric 101)
• Default Routes
  Both point at 10.0.4.1, but traffic prefers enP3p49s0 due to the lower metric.

Solution

Use Policy-Based Routing (PBR) to bind each source IP to its own routing table and interface, ensuring correct return-path selection.

Steps

1. Define Custom Routing Tables

   echo "100 rt_enP3p49s0" | sudo tee -a /etc/iproute2/rt_tables
   echo "101 rt_enP4p65s0" | sudo tee -a /etc/iproute2/rt_tables

2. Create Rules & Routes for enP3p49s0 (10.0.7.6)

   sudo ip rule add from 10.0.7.6 table rt_enP3p49s0
   sudo ip route add 10.0.4.0/22 dev enP3p49s0 src 10.0.7.6 table rt_enP3p49s0
   sudo ip route add default via 10.0.4.1 dev enP3p49s0 table rt_enP3p49s0

3. Create Rules & Routes for enP4p65s0 (10.0.7.5)

   sudo ip rule add from 10.0.7.5 table rt_enP4p65s0
   sudo ip route add 10.0.4.0/22 dev enP4p65s0 src 10.0.7.5 table rt_enP4p65s0
   sudo ip route add default via 10.0.4.1 dev enP4p65s0 table rt_enP4p65s0

4. Verify & Re-Test

   • Restart your iperf3 servers:

     iperf3 -s -B 10.0.7.6
     iperf3 -s -B 10.0.7.5

   • Run clients simultaneously and confirm ~940 Mbps on both flows.

---

By ensuring each source IP uses its own routing table, return traffic now follows the same interface it arrived on, restoring full-duplex TCP performance.

smb.conf

# /etc/samba/smb.conf
[global]
   server min protocol = SMB2
   server max protocol = SMB3
   socket options = TCP_NODELAY IPTOS_LOWDELAY SO_RCVBUF=262144 SO_SNDBUF=262144
   acl allow execute always = true
   acl map full control = yes
   deadtime = 60
   getwd cache = true
   min receivefile size = 16384
   strict sync = no
   sync always = no
   use sendfile = true
   aio write size = 1
 # Rest of config

sysctl.conf

# /etc/sysctl.conf
net.core.rmem_max           = 33554432
net.core.wmem_max           = 33554432
net.ipv4.tcp_rmem           = 4096 87380 33554432
net.ipv4.tcp_wmem           = 4096 87380 33554432
net.ipv4.tcp_window_scaling = 1
#vm.swappiness=10
#vm.vfs_cache_pressure=50