Link aggregation: Bolstering bandwidth

Getting two benefits from one technology is a network manager's dream. Link aggregation, a method for achieving both backbone redundancy and incremental bandwidth expansion, is now in the midst of an important transition from vendor-proprietary to standards-based implementation.

The emerging IEEE 802.3ad specification will deliver switch-to-switch and switch-to-server incremental bandwidth increases in a way that also brings inherent failover capabilities to Ethernet networks.

The specification can benefit organizations that want to add bandwidth to legacy 100M bit/ sec Fast Ethernet switches. More important, it will let organizations build high-bandwidth, resilient backbone networks using Gigabit Ethernet switches.

Without link aggregation, the Spanning Tree Protocol - used to guard against loops in Layer 2 internetwork bridges - prevents the addition of bandwidth. Spanning Tree only permits a single physical link to be active between any two stations, so it automatically puts the additional link into backup mode.

Link aggregation works by making two to six or more physical links appear as a single logical link to Spanning Tree and any other Layer 2 or 3 protocol. At the same time, link aggregation makes automatic failover possible by enabling the physical links to serve as redundant backups to one another.

With benefits like these, it's no wonder large network vendors have been delivering proprietary versions of link aggregation for some time, and that the IEEE standardization process is now underway with the 802.3ad task force.

The 802.3ad specification adds a link aggregation sublayer to the conventional Ethernet protocol stack at Open Systems Interconnection Layer 2, the media access control (MAC) layer. This sublayer effectively separates the physical connections below from the new, logical MAC address it shows to higher level protocols.

Within the sublayer, a link aggregation control protocol (LACP) performs functions that range from verifying configurations and operating status of participating devices to carrying out the distribution tasks necessary for assigning packet flows to their physical links. The LACP also carries out the collection tasks necessary for receiving incoming packets. Also, the protocol contains a control function for adding and deleting physical links.

The distribution mechanism determines which packet flows will go over which physical links. In the event of a link failure, the control function alerts the distributor, which then reassigns the packet flows. Because the operations are carried out low in the OSI protocol model, failure detection and reselection can occur very quickly, typically in less than a second.

A key distribution requirement of the 802.3ad specification is that, for Layer 2 devices, packet frame order be preserved for source-to-destination flows that require order preservation. This ensures that packets stay in sync for entire network conversations.

Layer 3 switches and routers, on the other hand, may elect to use their higher level intelligence to reorder packets as a means of improving network load balancing.

Value by vendor implementation

For example, it does not mandate the selection algorithm or protocol to carry out the distribution function. The selection algorithm is one of several areas in which vendors can add value through their own implementations.

Vendors will also be able to differentiate themselves by extending the technology to work not just with multiple links, but with multiple switches. This function will make it possible to add another dimension of resiliency to core backbone networks.

Of course, the beauty of link aggregation is that the inner workings are transparent to applications. Link aggregation thus gives network managers a relatively simple means of getting more bandwidth - and greater resiliency - out of existing topologies.

In switch-to-server applications, for example, multiple Fast Ethernet links could be added, along with redundant server network interface cards, to boost bandwidth and resiliency in a workgroup setting.

In a high-end configuration - for imaging or digital prepress applications, for example - four Fast Ethernet links could be aggregated from server to switch, for 400M bit/sec of bandwidth to the network. At the same time, this bandwidth could be extended through the switch-to-switch backbone, with six Gigabit Ethernet links aggregated between the high-speed switches for a whopping 6G bit/sec of resilient bandwidth.

Link aggregation is available today in vendor-developed implementations. What will be new is the technique's standardization via the 802.3ad specification. Technical work is expected to be completed before the end of 1999, and 802.3ad should reach formal ratification by March 2000.

Even today, link aggregation is a valid check-off item in new product evaluations because network managers can rely on the fact that major vendors will ensure their implementations are forward compatible with the coming standard.

Migration of Network Infrastructures to 10/100/1000Mbps Ethernet
Jato Technologies white paper that discusses high-speed networking in general and compares link aggregation to other approaches.

Link aggregation vs. fault tolerance
Network World Fusion Focus on High Speed LANs, 10/26/98.

Can Layer 3 switching increase bandwidth?
Compares Layer 3 switching with link aggregation. Network World Fusion Focus on High Speed LANs, 5/30/98.

Network World Fusion Focus on High Speed LANs
Archive for our free e-mail newsletter.

Pieper is a senior product manager in the Switching Systems Division at 3Com and Tolley is business development manager in 3Com's Advanced Technology Group. They can be reached at Karl_Pieper@3Com.com and Bruce_Tolley@3Com.com.