As mentioned in the last newsletter, we use the phrase cloud networking to refer to the functionality that has to be in the network and in the management of the network in order to support cloud computing. We are going to use this newsletter to discuss some of the changes that have to occur in the data center LAN in order to support cloud computing.
As mentioned in the last newsletter, we use the phrase cloud networking to refer to the functionality that has to be in the network and in the management of the network in order to support cloud computing. We are going to use this newsletter to discuss some of the changes that have to occur in the data center LAN in order to support cloud computing.
It is fair to say that over the last ten years that data center LAN design has been somewhat staid. For example, over the last decade most data center LANs were designed around a number of key technologies, requirements and design concepts. That includes the use of the spanning tree protocol at the link layer to ensure a loop-free topology; the use of Ethernet on a best-effort basis by which packets may be dropped when the network is busy; the need to support applications that are neither bandwidth intensive nor sensitive to latency; the use of switches with relatively low port densities; the use of a three tier LAN design and the separation of the data network from the storage network. Today, driven in large part by the need to support the creation and migration of a rapidly growing number of virtual machines, data center LAN design is far from staid. In particular, all of the previously mentioned technologies, requirements and design concepts are being seriously questioned.
As noted, the use of the spanning tree protocol to avoid loops is losing favor. The primary reason for that is that the spanning tree protocol prevents all available forwarding resources in a redundant network design from being simultaneously utilized. This dramatically limits the scalability of the LAN at a time when there is the growing requirement for data center LANs to be more scalable. It is now possible to ensure a loop-free topology without using the spanning tree protocol. This is accomplished by implementing a combination of switch virtualization and multi-chassis Link Aggregation (MC LAG).
In this context, switch virtualization refers to two or more physical switches being made to appear to other network elements as a single logical switch or virtual switch, with a single control plane. Link Aggregation is not new. For example, today it is common to create a high-speed uplink (a.k.a., a LAG) by aggregating multiple 10Gigabit Ethernet links. What MC LAG adds is that it allows the links of the LAG to span the multiple physical switches that comprise a virtual switch. From the server perspective, links to each of the physical members of a virtual access switch appear as a conventional LAG or teamed links. This means that switches can be virtualized without requiring any changes in the server domain. We will come back to the topic of the changing data center LAN several times in the next few months.
Jim recently completed an in-depth report on cloud networking. Given its length, we will publish the report over the next several weeks in five installments. We will also publish a complete copy of the report in mid-December. The first two documents have already been published.
