Transparent Interconnection of Lots of Links, or TRILL, is a connection-layer network standard recommended by the IETF. It holds significant importance because large data centers are beginning to leverage new technologies like Fibre Channel over Ethernet (FCoE) to converge storage and IP traffic onto Ethernet connections. The standard Spanning Tree Protocol (STP) will no longer be suitable for converged networks or the expansion of hyperscale data centers, which is where TRILL support comes in.
1. Previously, we commonly used STP. Its weakness lies in that it was designed in the era of ultra-small hubs, even predating the switch. Although STP has several variants, by design, it ensures only one path to a particular destination. The goal of STP is to create a loop-free topology. Of course, almost all networks have redundant paths.
With STP, all these redundant paths are blocked. As the environment grows, more routers and paths are added to the fabric, but STP still blocks all paths, leaving only a single path active.
2. When the active path fails, the network must reconverge on a new path. In large networks, this reconvergence process can take several seconds. While that might seem acceptable for standard IP traffic, it is unacceptable for storage or converged networks, especially those with virtualized environments. Another weakness of STP is that it is not very efficient from a network bandwidth perspective.
3. First, all blocked paths represent idle bandwidth. Increasing per-unit bandwidth means that more and more bandwidth is going unused. Second, the active path may not be the most efficient or shortest communication path between two devices.
In fact, data on STP often uses a “scenic route” over the network rather than the direct or shortest available path. This drawback not only impacts storage but is also detrimental to real-time live virtual machine migration in virtual environments. Migrating a VM or application to another server may require traversing several paths and switches, and suboptimal path selection only worsens performance. Virtual Machine (VM) migration also has to contend with other traffic on the primary path. In practice, many large virtual environments set up a dedicated VM migration network.
4. If those previously blocked paths were utilized effectively鈥攎eaning migrating VMs via paths idled by STP鈥攖hings would be much simpler. One goal of TRILL is to find the shortest available path and use it. Doing so requires understanding the entire topology and current network utilization at that moment.
Back when Spanning Tree was designed, hub/switch hardware couldn’t store the configuration state of the entire network. As a result, every path, whether active or inactive, had to be capable of handling peak loads. TRILL “knows” the entire fabric and how to utilize it efficiently, so the network can be used more effectively without requiring every component to handle peak loads.
5. TRILL essentially distributes network load across multiple paths, thus utilizing network bandwidth more efficiently. By adding multipath capabilities to Layer 2 networks, TRILL liberates network bandwidth and makes L2 networks more resilient and more suitable for virtualized environments. Previously, without TRILL, most networks were constrained by STP’s limitations and had to be built as multi-tiered networks: a Layer 2 architecture at the edge or access layer and a Layer 3 network at the aggregation layer. Finally, core routing protocols were set up on a different tier of the network architecture. This has been the dominant approach to network design for over a decade.
6. The idea behind this design is to partition the Layer 2 network into other tiers within the constraints of STP. This way, the recalculation process during a failure or traffic reconvergence can be kept within an acceptable time range. The downside of this approach is that this type of network is relatively expensive.
First, Layer 3 or routed ports are more expensive than Layer 2 switch ports. The more you deploy, the higher the cost of the architecture. The second drawback of introducing a Layer 3 network into the environment is its complexity, which requires ongoing tracking and management. For today’s increasingly specialized IT personnel, complexity is something to be avoided.
7. This design also makes it difficult for dynamic data centers to achieve on-demand service goals. With Layer 3 present, migrating bandwidth from a Layer 2 network to other network tiers requires meticulous planning, and flexibility is limited. Because of these factors, after