The data centre is the place where your valuable information lives. The data centre network allows you to access and process that data. It is obvious that this network needs to meet strict requirements with regards to reliability, security and throughput provided.
A decade ago, an architecture composed of 3 layers (core, aggregation, access) was common in the data centre network. Servers were often connected with only 1 Gbps and a different network was used for storage. Many workloads were running virtualized, but there were also quite a few still using bare-metal servers. Large layer 2 networks with spanning tree protocol and heavy oversubscription were very common. Many network providers sold proprietary data centre fabrics based on layer 2 technologies and often loosely based on TRILL.
The modern data centre network looks radically different compared to that old model. Today’s switches provide much higher port speeds. The current standard speed for server ports in 2021 is 25 Gbps. 100 Gbps is used as uplinks for switches between switches while 400 Gbps Ethernet is already available. Modern applications use microservices architectures and are often deployed with containers. This leads to much more traffic between servers (east-west). In addition, storage traffic has moved from Fibre Channel to IP and Ethernet to make use of the higher speeds provided by modern Ethernet switches.
This results in the old architecture with the heavy oversubscription and large layer 2 networks with a lot of BUM traffic (broadcast, unknown unicast, multicast) reaching its scaling limits.
Modern data centres are built as IP fabrics (routing between switches) in a closed topology which is also known as leaf-spine architecture. Since shared layer 2 segments between switches are often still a requirement (e.g., for VM migrations) traffic is encapsulated in VXLAN headers. This can be combined with the EVPN control plane which helps reduce flooding and makes the network more scalable.
While this solves many important problems, it also brings new protocols, and with it, new complexity into the data centre network. That is why orchestration and automation solutions are so important in the modern data centre.
Solutions can be self-built with tools like Ansible or Salt that use APIs from providers like Arista or Juniper, leaders in this field. This still requires a team to build up plenty of knowledge in completely new domains, like programming and automation.
Turnkey solutions like CloudVision from Arista and Apstra AIS from Juniper are here to take over all the heavy lifting and enable the network teams to focus on the real needs of their organizations, instead of being fast “CLI jockeys”.
The experts at ngworx can help you analyse the needs of your data centre network. With the knowledge gathered from implementing various data centre networks with different technologies and providers, we are well positioned to recommend and implement a solution that meets the needs of your organization.
Relevant Blog articles
