Managing datacenter network traffic the easy way

How the SONiC open-source ecosystem is helping build a connected networking fabric from core to edge to cloud

Sponsored The datacenter world is becoming more complex. In central facilities, performance requirements are increasing as service providers struggle to support more customers efficiently while driving down traffic latency. The evolution of edge computing is making infrastructure management even more challenging, as businesses demand more distributed storage and computing resources. So how can we simplify things?

Dell for one believes that a simple, flexible, approach to managing datacenter network traffic is key. And the company is releasing the latest version of its Enterprise SONiC network operating system with these challenges in mind.

The product is built upon Software for Open Networking in the Cloud (SONiC), an open-source Linux-based OS focused on network management in datacenter environments. Microsoft, which originally developed the system along with the Open Compute Project in 2016, gave it to the Linux Foundation last year. Since its launch, cloud providers and other Tier 2 service providers have used it to manage switch fabrics from multiple vendors in large-scale environments.

As with many open-source projects though, the initial concern for many enterprise SONiC users was support. Large operators need reliable help if something goes wrong. If a configuration issue arises, they need a partner which can offer guaranteed service level agreements (SLAs) rather than just forum posts. And if they discover bugs in a project, they won't always have the time or skill to fix the open-source code themselves in the necessary time window.

Enterprise SONiC bursts through

Dell responded to this need by launching the first commercial version of the SONiC product, Enterprise SONiC Distribution by Dell Technologies, in 2021. It was part of a journey that the company began in 2014 with its Open Networking initiative (ON). This sought to make datacenter networking more flexible for customers by separating the hardware and the software components of the network stack.

Dell also made many technical enhancements to the product. "We have done a lot of work on extending SONiC capabilities out to broader enterprise and telco use cases." says Saurabh Kapoor, director of Product Management and Strategy at Dell.

The company fleshed out some gaps in API support and added some new features. These included more standardized feature accessibility through its new CLI-based Management Framework, along with support for sFlow. The latter handles packet-and time-based traffic sampling for switches and routers and is useful for the kinds of high-speed networks found in cloud service provider environments.

Other enhancements included support for management virtual routing and forwarding (VRF). This subset of VRF technology separates out network management and data plane traffic. It sends out-of-band management traffic over its own virtual router so that the two traffic types don't mix. Users can now also enjoy streaming telemetry with SONiC.

Dell feeds its technical enhancements back to the open-source SONiC community, which then considers them for inclusion in the upstream product. "We have very active work streams and initiatives going on with the open source community." says Kapoor. Dell is on the governing board of the Linux Foundation, while also sitting on its technical steering and outreach committees to drive SONiC into new use cases.

The open-source approach is part of a broader widening of the Dell datacenter network management ecosystem. Dell signed with multiple partners to integrate their network management software with Enterprise SONiC for example, making it easier for customers to standardize on the software layer in their networks as part of a broader initiative to embrace software-defined datacenters.

The latest collaborations – which include BeyondEdge, Metalsoft and Racksnet - were forged to provide a choice of network management, automation and broader lifecycle management capabilities.

Embracing the edge

Edge computing has become a key focus for the Dell Enterprise SONiC team. More businesses across different sectors are mulling the idea of relocating selected software, data, and systems away from central datacenters to the edge of the network, driven by use cases for processing data closer to the point of creation or consumption. Whether you're handling cameras, PoS systems, or remote monitoring and maintenance of industrial devices, it sometimes pays to drive down data latency by hosting compute directly at factories, office buildings, or other facilities.

This represents a big potential investment for enterprises and service providers alike. IDC estimated a $176bn worldwide spend on edge computing last year, up 14.8% from 2021, and it will leap to almost $274bn by 2025, the company said.

If there's one common requirement among all of these sites, it's fast and reliable network connectivity, which is why Dell is now embracing the edge. It wants to help customers solve a range of technical challenges in edge computing and networking.

Companies must move large amounts of data and network infrastructure to these edge environments to manage data-hungry machine learning algorithms. They must harden communications between different devices to make them more secure, and need ways to manage digital infrastructure remotely.

Building a seamless continuum of network management between central datacenters and edge facilities was challenging, explains Kapoor.

"Edge environments are complex because you see a lot of diversity across use cases." he explains. Some edge engagements are dominated by small footprints and focus on ease of management. In others, security is a primary concern.

Dell launched its PowerSwitch E-series switch specifically to target these edge-based use cases. These provide up to 90 watts of Ethernet-based power per port to drive a range of connected devices in edge environments where power outlets are limited. They also include port security to ensure that the right devices are connected and to protect their communications.

The company has integrated the Enterprise SONiC software seamlessly to manage the E-series switches, making it easy to configure and troubleshoot them remotely where necessary.

A modular approach to network management was key to embracing edge connectivity as part of Enterprise SONiC's capabilities. A critical asset here is the software's container-based architecture, which enables operators to choose only features needed for their specific edge environment. Dell's Edge bundle supports lean CPU and memory requirements while also allowing customers to add their own in-house applications atop the network operating system. Dell Enterprise SONiC's user container manager feature also allows operators to dedicate the amount of CPU and memory that they want to allocate specific applications to make the most appropriate use of the hardware.

Taking Enterprise SONiC further

The first week of May 2023 saw a new release of Enterprise SONiC emerge, with version 4.1 sporting a new set of capabilities to help connect edge environments to cloud infrastructure.

One of the biggest features is a focus on open hardware. In the past year, Dell has been integrating with hardware vendors to expand its SONiC ecosystem still further. "Our customers wanted us to extend the value proposition we were delivering with Dell Enterprise SONiC on non-Dell platforms as well," Kapoor says, explaining that the company is driving hardware as a commodity layer of the datacenter technology stack.

"We work with these partners to test, validate, and qualify their platforms with Dell Enterprise SONiC," he adds. "So now our customers have access to all the test cases and the performance results for the specs we have tested."

These partnerships allow customers to build multi-vendor hardware stacks running the same operating system across the board, while still getting full support.

There are some other technical enhancements in the 4.1 release. One that will be especially interesting for service providers is 802.1Q-in-802.1Q (QinQ). This technology enables them to bundle end-user VLANs into single VLANs, which is useful for managing complex multi-tenant and large-scale network fabrics. It helps telco carriers to support more customer VLANs running over their infrastructure, offering more network segmentation and isolation options.

The other big announcement is RDMA over converged Ethernet (RoCE) v2. RoCE already allowed companies to access memory directly over an Ethernet connection by wrapping InfiniBand packets. Version two allows these sessions to be routed, so that they can occur across different Ethernet domains.

RoCE v2 will enable service providers to offer their customers higher performance for remote storage, making it look like local storage. The low latency and high-throughput will enable users to reduce or eliminate bottlenecks when accessing storage remotely. That makes it very useful in areas like high-performance compute, business analytics, AI, and real-time scenarios like financial trading.

This evolving software will help create high-performance fabrics with no vendor lock-in, says Kapoor, while providing end-to-end coverage out to the edge via one GUI that all partners can plug into. As service providers seek to tame increasingly complex infrastructure, having a single network operating system that provides visibility and centralized management across the entire fabric – from edge to core to cloud – is a significant advantage.

Sponsored by Dell.