M4300-24x24F LAG-LACP configuration issues

[hpc-encs]

Hi

I have 2 x M4300-24x24F stacked, and some servers connected directly to them via 10G Copper cables, 4 cables per server (2 cables per switch/unit).

My goal is to increase the bandwitdh using LAGs.

The servers NICs are configured in bonding: BONDING_OPTS="mode=802.3ad lacp_rate=fast xmit_hash_policy=layer2+3 miimon=100"

Ethtool detects: Speed: 40000Mb/s for the bonding interface (4 NICs)

On the switch side, I configured the LAG:

Admin mode:Enable

Link Trap:Enable

STP Mode:Enable

Static Mode: Disable

Hash Mode: Src/Dest MAC, VLAN Etype, incoming port

ports: 1/0/37,1/0/38, 2/0/37, 2/0/38

I tested the connection using iperf; and the result is always around 10Gbps.

Is something missing/bad in my configuration?

Thanks

Regards

[schumaku · Accepted solution]

Simply put: You need at least the same amount of sessions like you operate physical links. Each session will be run on a dedicated addresses. A common misunderstanding, no matter if people talk about IEEE standard LACP load balancing, or eg. Cisco proprietary implementations. 

[schumaku]

Hope you are aware even an LACP LAG does not magically create a single channel?

[hpc-encs]

Hi

It'd be much appreciated if you develop more your answer.

Thanks

[schumaku · Accepted solution]

Simply put: You need at least the same amount of sessions like you operate physical links. Each session will be run on a dedicated addresses. A common misunderstanding, no matter if people talk about IEEE standard LACP load balancing, or eg. Cisco proprietary implementations.