Link aggregation problem. Only one NIC registers traffic in report section

[acanessa]

Hi, thanks for taking the time to read my post.
I set up a vsphere 5 cluster. Two are connected to the freenas 8 server via switch.

The hardware:
I have two TP-LINK TG-4368 NICS (Chipset Realtek 8168) in the freeNAS 8 server. Plus a 3Com 100mbit for WEBgui. I have the same two NICS in each ESX 5. All the NICS go to a 3com Switch 3CRBSG2893, it's about 500 USD and its manageable. You can use LACP in it.

The problem:
When I use the lagg0 to do vMotion or copy large datastores, the speed is not near 2GB (2 x 1GB links), and i see in the reports section that the re0 nic is about 600Mbit tops but the re1 NIC has basicaly no traffic.

What I am doing wrong here? Shouldn't both NICS register half the overall traffic? I did set up the LACP in the switch, for the ports where the freenas nics are attached.

I have tried load balancing too, same problem.

Any ideas?

Thanks!!
Augusto

 

[acanessa]

For some reason the images came out too small. :S

 

[b1ghen]

Any single session can only go over one NIC, so LAGG only makes a difference when you have multiple sessions going on simultaneously. You will never see more than 1Gbit in one session.

 

[jrodder]

I experienced this same thing. I was testing with LACP, and only saw the one NIC traffic. OK, so I understand that it's only one session. But, isn't LACP also fault tolerant? I tried testing the LAGG setup by pulling cables, and I lost ping to the LAGG IP when I disconnected the NIC that was showing my traffic.

 

[jamoses]

acanessa, jrodder and others,

Were any of you able to get LACP working reliably or at all? Please check out my posting "LACP Load Balancing NOT working
". I have 4 XenServers, each with a 1GbE NIC. LAGG, LACP set up on 4 cards just does not work. All traffic goes through one port. The LAGG is correctly "partnered" with the switch. The switch properly reports the MAC of the the four LAG ports as the NIC of the priority port. Failover works great, but I need throughput.