Topologies used for fault tolerance

Network topology is the logical arrangement of communication devices and resources in the computer network. It is built to develop a standardized structure of communication between the servers and clients and define the hierarchies for the data flow. Following are some of the commonly known network topologies used to define data flow.

Network topologies are vital in ensuring that the system's network is fault-tolerant and provides reliability. To achieve fault tolerance, these topologies are further modified to form different data center topologies that are more efficient and adopt redundancy, replication, and load-balancing approaches to obtain a fault-tolerant network. Let's look at the major ones where each proceeding one is better than the previous one.

Fat tree topology

It is widely used for the clustering of cloud data centers and high-performance computing. In this topology, nodes are organized into multiple stages that enable bidirectional communication between the nodes. Data passes through intermediate nodes at each stage before reaching the destination. This provides efficient communication and data transfer among multiple interconnected nodes.

It has a multi-tiered design with three main layers:

• Core layer: It has high-performance switches that are considered the network's backbone and provide connectivity between the aggregation layer switches.

• Aggregation layer: It has switches that efficiently aggregate traffic from the core layer to the edge layer.

• Edge layer: It consists of the switches that are connected to the servers where each switch is assigned to a unique server.

Clos network topology

It is a multistage networka network for interconnecting a set of nodes through a switching fabric. that provides a highly inter-connected network through a switches architecture that reduces the total required ports for making a connection. It has a non-blocking architecture in which all the inputs can be connected to any available output without causing any congestion and having sufficient alternate paths in case of any failure.

It contains three stages, each of which is prepared using crossbar switches:

• Ingress: Receives incoming data or traffic from external sources.

• Middle: Consists of multiple interconnected switches or nodes that perform the main data processing and switching functions.

• Egress: Prepared for delivery to its final destination.

BCube topology

It is a recursive server-centric network designed to connect many servers while maintaining the fault-tolerance in a cost-efficient method. The recursive nature implies that the network can be extended easily by adding more servers and switches. Moreover, its distributed nature improves fault tolerance because if one switch fails, the other can substitute instantly.

In this network, the servers are divided into sets, with n servers in each set, i.e., 4 in this example. Those servers are then connected to the switches, one set to one switch, and there are a total of n switches in the middle layer. Then there are additional n switches in the top layer, and each server from a set is connected to a unique switch.

DCellDistributed Data Center in Cell topology

It is also a recursive network topology mainly designed for large-scale data centers to connect the servers and switches in a fault-tolerant environment. It has redundant connections and can easily accommodate the growing amount of data in the network, making it a suitable option for managing high computational workloads. Moreover, It implements a decentralized approach by keeping the routing decisions and communication within the cell to improve scalability.

The main component is DCell0, which contains n number of servers and a single n-port switch to which every server in DCell0 is directly connected. There are multiple other cells containing the same network structure that are interconnected and form a multi-level hierarchical structure. These connections between the cells provide alternate data flow routes if one fails to improve fault tolerance.

Summary

When creating a fault-tolerant environment, it is important to develop a structure that provides high-level reliability for complex workflows. Therefore, the network topology designs should be carefully evaluated and then adopted. Fat tree topology, Clos network topology, BCube topology, and DCell topology are some of the main topology networks that have been structured to achieve fault tolerance in cloud computing.

Test your understanding