Link Aggregation

Link aggregation is a common term in the IEEE from IEEE 802 range (more precisely, 802.3ad, since 2008 802.1AX ) and refers to a method for combining multiple physical LAN interfaces into a single logical channel. The first implementations of this technique are from Kalpana (now Cisco) and originally were used solely to increase the data throughput between two Ethernet switches ( also invented by Kalpana ). Current implementations can, however, also connect servers and other systems via a link aggregation.

Other names

Although there are incompatibilities and technical differences (depending on the manufacturer or context) used for the bundling of Ethernet interfaces, the following terms as synonyms:

  • Bonding, in the Linux environment.
  • Etherchannel, Cisco.
  • Link aggregation, IEEE.
  • Load Balancing, general.
  • Port aggregation, at Hewlett -Packard.
  • Trunking, Brocade, Sun Microsystems and several other manufacturers.
  • Bundling, of course, there is also a German term for the Ethernet channel bundling.
  • Teaming, Novell Netware
  • Bridging, at Microsoft

Operation

Generally more than one physical Ethernet interfaces to connect two devices are connected together, that is, two interfaces can be bundled to transport the double amount of data, therefore, four of the four -fold amount. It makes sense to combine fast as possible interfaces is twice Ethernet (half duplex ) correspond to 20 Mbit / s, three Fast Ethernet (full-duplex ) and Fast Etherchannel are already 600 Mbit / s A maximum increase in performance is achieved for many bundling method, when the number of bundled ports is a power of two. For example, almost the bandwidth of a Gigabit Ethernet interface can be achieved by aggregation of eight Fast Ethernet interfaces. Another advantage is increased resiliency. As can vary one or more interfaces, without the logical channel is interrupted in some processes. Only the data throughput decreases according to the missing links.

Bundling process

  • Round Robin: Here all available lines are alternately sequentially used.
  • DA trunking: Here you select the basic interface based on the modulo of the destination MAC address.
  • SA trunking: Here you select the basic interface based on the modulo of the source MAC address.
  • SA -DA - trunking: Here you select the basic interface based on the modulo of the source MAC address and destination MAC address.
  • Adaptive Trunking: Here is another only switched on at 100 % utilization of the first elementary interface.
  • Dynamic trunking: Using the proprietary PAgPs, or the IEEE -compliant LACP can be defined dynamic trunks.
  • Further methods.

Basic implementation

The basic implementation of a channel bundling between two Ethernet switches come with surprisingly small changes (compared to a normal switch). Of course, the management interface must be extended to define the trunks. Then the way, like a switch gets its SAT (English Source Address Table - table with the MAC addresses of the sender) affected learning, and finally, broadcasts and packets that are addressed to unknown destination MAC addresses from trunk treated separately.

  • Management Interface: that the menu structures are concerned, certainly needs no great explanation.
  • Address learning phase: Receives a switch on a trunk port, an unknown sender address, so these are not automatically the address stack ( SAT) is assigned to this port, but the switch tries to match all the members of the trunk a same network load and therefore new distribute sender addresses evenly on all companies in the trunk ports. Thus, for example receives the port that the fewest entries leads currently in its satellite, the new address. Depending on the implementation, other distribution strategies are used, see Section bundling process.
  • Broadcasts and unknown addresses: These are not sent in a trunk on all ports, but only one per trunk line, generally the port is selected with the lowest port number. Depending on the implementation, other strategies are used to select the port, see section bundling process.

Interestingly, no further action is required to implement a simple implementation. The implementation of the transport mechanisms (cut through, store and forward, etc. ) does not change. Nevertheless, this basic implementation is far from optimal and is really only for the coupling of two nets. A disadvantage is that each MAC address only one port of the trunk is used - for a single station, it means that she has no benefits - benefits only the network as a whole. Further, it may happen that the most active stations are randomly assigned to all the same port, hence the load is not evenly distributed within the trunk. There are also bundling method, which can bind powerful servers. All these facts have led to many manufacturers offer different, but also more advanced implementations. These are not generally compatible with each other. These incompatibilities have meant that the IEEE has addressed the issue and LACP (IEEE 802.3ad ) has specified.

Institute of Electrical and Electronics Engineers Hewlett-Packard Port Aggregation Protocol Broadcasting (networking)
15646
de