In this video I walk through bonding two 1GbE NICs on Oracle Linux 8.6 (basically RHEL 8) so I can push files off a small server faster—think 100GB chunks going up to my big storage box. The goal is a 2Gbps uplink back to my HPE 2920 switch, and I specifically use round-robin (mode 0 / balance-rr) because it’s the one bonding mode that can stripe packets across both links for a single TCP/IP flow. I do the whole config using nmcli, starting clean (no existing connection profiles), creating bond0, then adding eno1 and eno2 as slaves, and verifying link speed and MAC behavior. Then I show the part people skip: the switch side. If you don’t configure a trunk/LAG, you’ll get MAC flapping—your MAC bounces between ports and performance gets weird. On the 2920 I build a standard trunk (not LACP), confirm the MAC lands on the trunk interface, and then I validate performance with rsync and iperf3. The big takeaway from the iperf testing is that most switches won’t do true round-robin load balancing; they hash. With the 2920, changing trunk load balancing from L3 to L4 makes a huge difference for multiple parallel streams (iperf -P), but it still explains why bonding is very directional and why LACP is usually the “correct” approach when you’re not specifically trying to solve this exact upload/offload scenario.