LACP between Unix-AIX and procurve switch.

[stephnane]

Hi.

I want to configure LACP between an HP procurve switch (5412zl) and an AIX server with 2 x 1Gb links.

Here are the result of our tests :

1. Send data from this server to 2 others servers -> seems to have a real aggregation, with a 2Gbps bandwidth.
2. Send data from 2 (ore more) others servers to this one -> seems to have only 1 Gbps bandwidth of network activity on the server.

Have you any idea to have 2 Gbps in both ways ? Does LACP support the aggregation in both ways ? Is there any other way to have 2 Gbps bandwidth ?

Tks.

[John Gelten]

Ultimately, when a packet has to be transmitted over the link, the sending device (in your example both the switch and the server) has to make a decision to send it over one of the available physical interfaces. This decision is not made by flipping  coin, but it is actually calculated based on very strict rules. This decision can be made on the MAC address, the IP-address or the portnumber of either the source or the destination host; or a combination of these.

Usually, by default the decision is made based on a combination of source and destination MAC address. In many switches you can change this load-balancing hash to better suit your needs.

You can imagine if your equipment is doing the load-balancing based on only the destination IP- or MAC-address, you will by definition only use one of the links for all your traffic that is going towards the server. Traffic from the server will usually be loadbalanced, because of the higher number of destinations.

Every now and then, I am involved in a project where the topology is so 'funny' that trunking does not reliably create the neccessary bandwidth, so an upgrade to physical 10Gbps is needed. For example, when you need enough bandwidth to transfer lots of data from a large fileserver to a backupserver every night, as fast as possible. In that case both of the MAC- and IP-addresses are the same, as is are the portnumbers, so the hash will always resolve to the same physical interface within the trunk.

Both LACP and the 'trunk' type support use of the total aggregate bandwitdh in both directions,provided you are using the appropriate hashing in both directions that are suitable for your specific environment.

This explanation is valid for all trunking/aggregation/channeling technologies in all brands that bundle more links into one bigger-bandwitdh logical interface...

For your HP E5400 series switches, see the Network Management and Configuration Guide and find the Outbound Traffic Distribution Across Trunked Links section at page 4-28 for more explanatory details and the information on how to change the hashing algorythm in this switch.

As far as I know, this switch also includes the source into the calculations, so if you test this with a more statistically relevant number of hosts, you will probably find that both links are being used.