Introduction to the routing transparency protocol (TRILL)

The transparent protocol is the connection layer network standard recommended by IETF, or TRILL for short. It is of great importance, because large data centers begin to use new technologies such as Ethernet fiber channels to integrate storage transmission and IP Address Transmission to Ethernet connections, and the standard Spanning Tree Protocol (STP) it will no longer be suitable for the expansion of converged networks or ultra-large data centers, which requires transparent protocols.

I. We used STPS in the past. Its weakness is that it was designed in the era of ultra-small hubs, and even before the switch. Although STP has several types, it is designed to ensure that there is only one path for the connection to a specific point. The goal of STP is to create a loop-free architecture. Of course, almost all networks have redundant paths.

Through STP, all these redundant paths are blocked. As the environment expands, multiple routers and more paths are added to the architecture, but STP still blocks all paths and only leaves one path.

2. When an active path fails, the network must be re-converged on the new path. In a large network, the re-convergence process takes several seconds. Although this seems acceptable for standard IP communication, it is unacceptable for storage or converged networks, especially those with virtual environments, another weakness of STP is that STP is not very efficient from the perspective of network bandwidth.

3. First, all blocked paths represent idle bandwidth. The increase in bandwidth per unit means that more and more bandwidths are not available. Secondly, the active path may not be the most effective or shortest communication path between two devices.

In fact, STP data often uses "Beautiful paths" on the network rather than direct or shortest available paths. This disadvantage will not only affect storage, but also affect real-time virtual machine migration in virtual environments. To migrate a virtual machine or application to another server, you may need several paths and switches. The sub-optimal path selection will only make the performance worse. Virtual Machine (VM) Migration also needs to compete with other transmissions on the primary path. In fact, many large virtual environments have a dedicated VM migration network.

4. If you make good use of the aforementioned blocked path, that is, migrating the virtual machine through the previously idle path of STP, the process will be much simpler, one goal of TRILL is to find and use the shortest available path. To achieve this, you need to understand the entire topology and current network utilization.

In the generation tree design era, the hardware of the hub/switch cannot store the settings of the entire network. The result is that each path, whether active or inactive, must be able to handle peak load. TRILL "knows" the entire architecture and knows how to effectively use this architecture, so the network can be used more effectively without the need for each unit to handle peak loads.

5. TRILL actually splits the network load into multiple paths to make more effective use of the network bandwidth. By adding the multi-path function to the L2 network, TRILL frees up the network bandwidth and makes the L2 network more elastic and more suitable for the virtualization environment, most networks are limited by STP restrictions and must build multi-layer networks, that is, the Layer 2 architecture of the edge or access layer and the layer 3 network of the aggregation layer. Finally, set the Core routing protocol on the other layer of the network architecture. This is the main method of network design over the past decade.

6. The idea of this design is to partition the second-layer network to another layer within the limits of STP. In this way, the re-computing process can be kept within an acceptable time range when a fault occurs or a re-converged transmission is required. The disadvantage of this method is that the cost of this type of network is relatively high.

First, the layer-3 or route port is more expensive than the layer-2 switch port. The more you deploy, the higher the architecture cost. The second disadvantage of introducing the layer-3 network in the environment is that it is complex and requires continuous tracking and management. Complexity should be avoided for IT staff who are increasingly specialized in division of labor.

7. This design also makes it difficult for a dynamic data center to achieve on-demand services. In the case of layer-3, it is necessary to carefully plan and limit the flexibility to migrate bandwidth from layer-2 to other networks. Because of this, the layer-3 network can only be deployed on a relatively small scale, and in most cases, the data center needs to face STP inefficiency.

As the converged network continues to develop and expand, the restrictions on the spanning tree will become more and more prominent, and the time required for re-convergence of connections will become a greater challenge, especially for the storage or virtual server architecture. FCoE requires lossless Ethernet transmission.

8. Enhanced Ethernet and DCB can provide lossless Ethernet. In addition to the L2 multi-path function, TRILL also makes it possible to enable the multi-hop FCoE. Therefore, through TRILL, administrators and engineers have the opportunity to adopt more complex technologies in the data center, as FCoE market penetration increases, enterprise storage will begin to join other protocols on the IP network. The combination of FCoE and DCB improves the efficiency of the architecture, and the number of endpoints they can connect can be at least doubled.

The TRILL virtualization technology can make the environment more dynamic. All in all, network convergence makes the data center larger and more elastic, which is also the result we want. We believe that few people are considering network transformation, as long as you make necessary preparations, then it will succeed.