• No results found

What Is Link Aggregation?

While something like STP is necessary to prevent network loops, it sure seems like a shame to have perfectly good network connections sit idle, just biding their time waiting for a link failure that might never come. We can make these links useful without angering STP by using link aggregation.

Link Aggregation 37

Link aggregation is the act of bundling multiple physical ports into a single logical group-ing. This prevents any one member of the group from being blocked, as the group appears as one logical port to STP.

Link aggregation offers a couple of advantages over discrete links managed by STP:

Q Increased Bandwidth and Load Sharing: Because multiple links are active within a logical group, additional bandwidth is available among unique sessions of traffic. Per Figure 4.7, no single traffic session, such as a single client talking to a single server, can use more than one physical link. Think of it like making a large highway with multiple lanes: You can fit more cars on the highway, but each car can only take up one lane at a time.

Figure 4.7 A data flow with link aggregation from a single source to a single destination travels through only one aggregated port.

Q Improved Redundancy and Availability: If a single link in the group fails, other links continue to pass traffic and the group does not go down. There is no need for the STP topology to reconfigure itself or for STP to transition a new link from blocking to active.

One of the challenges with link aggregation is that it comes in a number of names, shapes, and standards. Terminology varies between vendor implementations, even when imple-menting the same standards. To un-muddle this, we try to use the generic term Link Aggregation Group , or LAG, to describe the general concept, and use vendor-specific terms when talking about their specific implementations.

38 CHAPTER 4 Advanced Layer 2

Implementation Methods

There are many different ways to build a LAG, as each vendor decided to use a slightly different name or method in order to accomplish a logical link grouping. This can cause confusion and frustration to networking newcomers, so we cover the more common terms and technologies used.

802.3ad and 802.1ax – IEEE Open Standard

The IEEE LAN/MAN Standards Committee sponsored link aggregation in their 802.3ad open standard. The idea was to meet a long list of goals for link aggregation through non-proprietary means. Later, the standard was formally transferred and published as 802.1ax to avoid some confusion with other 802.1 layers. The formal definition of 802.1ax is as follows:

Link aggregation allows one or more links to be aggregated together to form a link aggregation group, such that a media access control (MAC) client can treat the link aggregation group as if it were a single link.1

Within the 802.1ax open standard, the IEEE also defines the Link Aggregation Control Protocol (LACP) . The purpose of this protocol is to allow two systems to negotiate a LAG using a standard, nonproprietary protocol.

EtherChannel – Cisco

EtherChannel is a Cisco proprietary way of building a LAG between Cisco switches. It can consist of up to 8 active ports to forward traffic and 8 inactive ports, sometimes called failover ports , to take over for any active ports that happen to fail. EtherChannel comes in two flavors:

Q A Static EtherChannel , or “mode on” in Cisco IOS-speak, is manually configured and will not use a negotiation protocol to build the LAG. If the network administrator inputs the wrong information in the switch, the LAG might still appear to be active but might not forward traffic properly.

Q A Dynamic EtherChannel can use one of two protocols to automatically create a LAG:

Port Aggregation Protocol (PAgP) or LACP. When a dynamic EtherChannel is cre-ated, both network switches involved negotiate to form the LAG. If the negotiation fails, the LAG is not established.

Link Aggregation 39

Etherchannel Versus Port Channel

Further complicating any discussion on link aggregation, the terms EtherChannel and Port Channel are often used interchangeably. There is a subtle difference, though. When config-uring an EtherChannel between two switches, you create a Port Channel interface on each switch, bundling the physical interfaces together. An EtherChannel, then, is the logical pipe between switches consisting of a bundle of cables, while a Port Channel is the logical interface terminating the EtherChannel at each end consisting of a bundle of interfaces. Put another way, when you make a configuration change to EtherChannel, you do so via the Port Channel’s interface. And when these changes are committed, they are automatically applied to each of the interfaces from which the EtherChannel has been constructed.

Other Vendor Terminology

Here is some additional terminology you are likely to encounter when discussing link aggregation:

Q Trunk (HP): Some documentation for HP network equipment (and many network engineers that used to support said equipment) refers to a LAG as a trunk. This can cause confusion between LAGs and links carrying 802.1Q VLAN tags, as the latter is termed a trunk by just about everyone else. The HP “trunk” implementation con-forms to the 802.1ad standard which is what non-HP engineers would call a LAG.

Q Multi-Chassis Link Aggregation (Various) or Cross-Stack EtherChannel (Cisco): A multi-chassis LAG has terminating interfaces on two different nodes. This is a more highly available implementation, as the LAG can sustain the loss of a connected tar-get. Implementation details are vendor-specific and outside the scope of this book, but you should know that MC-LAG exists and that it is more robust because the loss of a single switch chassis doesn’t mean the loss of the aggregated link. Examples include virtual port channel (Cisco) and distributed trunking (HP).

Q NIC Teaming: Many endpoint operating systems and hypervisors are capable of logi-cally bundling network connections. This is typilogi-cally referred to as NIC teaming or NIC bonding . In these configurations, a software driver must determine how to distribute traffic across the uplinks, as the connected physical switches are usually unaware of the teamed configuration.