LAG between SLM2048 and SG300-52 can’t exceed 1Gb/s

[1099pro1099]

We are trying to create a 4 member (4Gb/s) static LAG group between two switches, a Cisco SLM2048 and a Cisco SG300-52. The LAG group (4 x 1Gb) shows up on both switch and we do get 1Gb/s of throughput between the switch but that’s it. Testing with a flat network and 2 x 1Gb computers on each switch shows we cannot exceed 1Gb/s of combined throughput through the 4 member LAG. Pulling out any 3 of the 4 members shows that the throughput stays unchanged at a maximum of 1Gb/s . Putting all 4 computers into one switch shows they can do a combined 2Gb/s total. So why can’t the 4 member (4 x 1Gb) LAG group exceed 1Gb total?

[David Hornstein]

Hi Jean,

You are forgotting something about LAG, especially when you test the functionality.  I have taken the liberty to copy and paste a section from page 89 of the 300 series admin guide, which discusses packet distribution in a LAG and load balancing;

Load Balancing

Traffic forwarded to a LAG is load-balanced across the active member ports, thus achieving an effective bandwidth close to the aggregate bandwidth of all the active member ports of the LAG.

Traffic load balancing over the active member ports of a LAG is managed by a hash-based distribution function that distributes Unicast traffic based on Layer 2 or Layer 3 packet header information. Multicast packets behave in the same way as Unicast packets.

The switch support two modes of load balancing:

• By MAC Addresses—Based on the destination and source MAC addresses of all packets.

• By IP and MAC Addresses—Based on the destination and source IP addresses for IP packets, and destination and source MAC addresses for non-IP packets.

So,  if you unicast from one IP host to another  IP Host, LAG will decide which link (not links) in a LAG group to send that traffic over.

If you send Traffic from one source IP address to many destination IP addresses, then the switch LAG hashing algorithm can start to distribute traffic over multiple links in a LAG group.

So,  a web server that serves many remote IP hosts (destination IP addresses)  on a LAN,  could  start to evenly  distribute traffic over a LAG more efficiently.  The more remote hosts,  the more evenly the HASHing algorithm will start to evenly spread the traffic over the interfaces in a LAG group..

You will find,  that the more IP hosts or ethernet hosts on a network the more evenly traffic will start to distribute over a LAG group.  But in testing from one IP Host to another the switch will send traffic over only one link, until that link fails.  If that link fails you would observe the traffic will be sent over another member of the LAG group.. 

But you may be unlucky enough to observe, depending on the switch HASHing algorithm, to see one link in a LAG group be used more frequently within a LAG group.

See the following  IEEE link for a summary of how 802.3ad works.  http://www.ieee802.org/3/hssg/public/apr07/frazier_01_0407.pdf 

it also mentions;

"All packets associated with a given “conversation” are transmitted on the same link to prevent mis-ordering"

regards Dave