Data centre networking: SDDC

This blog post is part of our data centre networking series:

• Data centre networking : What is SDN
• Data centre networking : SDN fundamentals
• Data centre networking : SDDC
• Data centre networking : OVS and OVN
• Data centre networking : SmartNICs

In the previous blogs, we covered the architecture and main drivers behind software-defined networking. In this one, we discuss the impact of softwarisation on the other important data centre building blocks, culminating in software-defined data centres (SDDC). SDDC occupies a progressively larger segment of the cloud computing space, originally adopted by public cloud service providers and hyperscalers, and now finding a home with private cloud service providers, too. First, let’s consider which drivers influenced the evolution of data centres.

In the early days, IT applications were monolithic, and everything was built in a single chassis. Then information systems started progressively adopting a 3-tier architecture:

• Front End
• Back End
• Database

These 3 blocks needed a network infrastructure to communicate with each other.

The trend was then to break down the application servers into several independent components, which are:

• Compute (computing capacity)
• Storage

The traffic between the CPU and disks, which was carried on the internal bus of each machine in the past, is now on the network.

Virtualising physical servers among multiple logical units increased the bandwidth requirements from each physical unit attached to the network.
The latest trend is to virtualise network functions and disaggregate the physical components that constitute it. This has further increased the pressure on data centre infrastructures.

Virtualisation’s impact

With the increasing usage of virtualisation techniques and the “disaggregation” of network components, traffic within the data centre (known as East-West) has highly increased.

Data centre network architectures that were initially designed to meet “north-south” traffic needs proved unsuitable. Access bandwidth kept increasing dramatically (10G, 40G, 100G, 400G) to meet multiple instances on a single physical base.

With the routing being carried out at a”high” level in the infrastructure, we observed that the traffic exchanged between applications within the same data centre soon saturated the north-south links. Almost all the traffic, even local, goes up to the “core” routers without leaving the data centre.

In order to respond to this increase in traffic within the data centre, a new fabric architecture was implemented, known as “CLOS fabric.” It is based on new protocols, eliminating Spanning-Tree and optimising the use of all links.

On the other hand, the hierarchical “pyramidal” structure of the traditional data centre networks had to be replaced by a much flatter structure. This type of architecture is called “Spine and Leaf”.

Thanks to the power of virtualised compute and storage and SDN, it becomes possible for customers to subscribe to complete data centre infrastructures. Vendors no longer just provide storage and server performance, but also offer network components (also known as VNFs) such as switches, load balancers or firewalls as virtual resources that software can easily control. It’s no surprise, then, that not only vendors, but also a growing number of experts consider such a software-defined data center to be the data centre model of the future. The following sections explain why this is the case by walking through how a software-driven network works.

What is SDDC?

Software virtualisation and the easy provisioning of software-defined computing and storage have long been established in the portfolio of cloud computing providers. For some time now, the futuristic IaaS model, in which higher-level software acts as the central control unit, has also been used for the management of network infrastructures (SDN). This centralised control should ultimately reflect a model-driven approach from the network to compute and storage.

An SDDC combines these three application-driven infrastructure services in addition to an orchestration and management layer so that users can build their own custom data centre. Using the appropriate software, typical components such as routers, switches, server load balancers or firewalls can be provisioned and individually organized in the virtual network without the need to purchase dedicated hardware. Usually, the IaaS providers take care of it, which is why they are also responsible for the maintenance and security of the devices. If certain components are no longer needed, they can be removed from the SDDC at any time. 

The most recent forms of SDDC also reflect the impact of application disaggregation into microservices and the upcoming of container machines. Containers are deployed in clusters and managed using different Kubernetes technologies.

How does all this work?

As with all software-defined IT resources, the separation of control and data planes is an important characteristic of software-defined data centres. All control functions are extracted from the integrated hardware and implemented in a higher level software, which then functions as the virtual data centre’s seat of command. It takes care of all the tasks beyond simple data processing at the “data plane” level, for example, determining where an application or a certain process is executed, which path a data packet should take, or where files should be stored.

Communication with the different devices leverages standardized protocols such as OpenFlow, BGP or NETCONF. This allows the control plane to uniformly control and extend the functions of hardware components, regardless of their original manufacturer. Furthermore, there is no need to make changes on each device individually, because the corresponding information can also be transmitted centrally, either in an automated way or with a few clicks.

What advantages can SDDC provide?

Some of the key goals behind software-defined services is maximum flexibility, automation and efficiency. Achieving these goals requires overcoming hardware dependency and achieving a maximum degree of virtualisation. The result is IT infrastructures without the physical limitations of traditional environments, which can be flexibly tailored to user needs and scaled accordingly. After virtualising mass storage, server and networking solutions, this approach has then been extended to a complete virtual data centre in the form of a software-defined data centre. This has shown several advantages over traditional data centres with dedicated hardware:

• High reliability: SDDC solutions are very secure because the software architecture easily compensates for hardware failures. To accomplish this, the corresponding workloads are transferred to other components in a very short time, without the need for manual access to the failing devices.
• Enhanced Security: Unlike traditional data centre infrastructure, the SDDC model allows all security-related information to be tied to the virtual machines themselves. Security breaches, for example due to inadequate configurations, can therefore be practically ruled out. There are also some software-defined security solutions that help implement a granular microsegmentation, reducing east-west threat propagation.
• High planning reliability: Using a software-defined data centre equips IT infrastructures well for future technological developments where new functions can be integrated much more easily. The right scalability also helps to increase or decrease the selected resources at any time.
• Simplified hardware management: SDDC reduces isolated hardware technologies and enables the centralized management of all onboarded components. This makes it easier for providers to create and deploy resources, and for customers to control and monitor them. Enterprise customers also benefit from eliminating the need to buy their own equipment.
• Lower Costs: SDDC can be more cost effective than traditional data centres for two reasons. On the one hand, much of the hardware is cheaper due to the lack of control logic. On the other hand, simple scalability allows customers to keep costs as low as possible.
• Less vendor specific know-how needed: Standardization, open protocols and manufacturer-independent programming mean not only more flexibility, but also less reliance on specialist knowledge.

What are SDDC’s challenges?

Software-defined data centres take full advantage of virtualisation technologies, potentially making administration a daunting task. Virtual components and the underlying hardware should form a unit that is easy to manage and monitor. In particular, mobility and the fluidity of borders between physical and virtual data centres represent major challenges for operators. For example, the tools for mediating between the virtual and physical environments must be perfectly coordinated. Otherwise, a virtual component can report an error even if it originates from the physical level.

Combining different software-defined services (compute, storage, networking) is also more difficult than it might look. The different software-defined IaaS offerings are of varying maturity and might support various protocols and interfaces, which also need to be combined in the SDDC. We can see this reflected at the hardware level – which, on one hand, should be easy to provision and virtualise, but on the other hand, should support the interaction of various virtual server, storage and network components beyond its own material limits. This is the only way for the software-defined data centre to guarantee a multi-tenant usage of the infrastructure.

Fortunately, Canonical has developed a solution to these challenges. Canonical’s Charmed OpenStack is a complete suite which offers model-driven orchestration and management for modern applications and infrastructure deployed in open data centres.

All what you need to start using OpenStack

Canonical’s Charmed OpenStack is a pure upstream OpenStack distribution engineered for the best price-performance with full enterprise support. Learn more >

Ubuntu is the most popular Linux distribution across public clouds, data centres and the edge with a built-in security and compliance for enterprise customers. Learn more >

What is next?

The next blog will be dedicated to OVS and OVN as part of the open data centre networking stack.

The next posts in this series will cover:

• Data centre networking : OVS and OVN
• Data centre networking : SmartNICs