I’ve been watching Cumulus Networks pretty much since it was founded. The company is all about re-thinking how networking is done for large enterprises and its approach flies in the face of many decades of networking vendors’ behavior. Until recently, networking devices were veritable black boxes – sealed hardware devices that came bundled with completely proprietary software. As would be expected, these offerings from the likes of Cisco and Juniper were inflexible and expensive and gave customers very little flexibility to adjust to changing operating situations.
Which is where Cumulus comes in. The company created an open source networking software solution that allows customers to use standard hardware components. More recently this software-only approach has been modified and Cumulus now offers their own switches, but with the same open and flexible ethos that they’ve had since their inception. Cumulus’ vision hasn’t changed – they’re focused on unlocking the vertical network stacks of the modern data center.
And, as most people know, the stacks that modern organizations are using have been changing of late, and increasingly we’re seeing these organizations experimenting with containers and microservices, both of which introduce new challenges and opportunities for organizations dealing with the compute, storage and networking aspects of infrastructure. To this end, Cumulus is today announcing its Host Pack offering, a suite of technologies that help operators wrangle both visibility and connectivity within the context of containers and microservices.
And don’t underestimate just how much container usage changes the game for networking. Because containers are constantly created and destroyed, and workloads are often moved to different physical machines or migrated to completely different data centers, a new way of thinking about the underlying network is essential to success. Cumulus’ offering is attempting to address the challenges network operators face in achieving end-to-end network visibility and connectivity of containerized applications. Some specific issues that operators need to resolve include:
- Network Blind Spots: The short-lived, ephemeral nature of containers makes them difficult to identify and track, making consistent, real-time visibility and troubleshooting with traditional network tools virtually impossible. The accessibility of containers also makes it possible to put them on untrusted network segments, leading to dangerous network blind spots.
- Increased Network Complexity: The fleeting nature of containers and microservices demand a simple yet high-performing network that adapts quickly to the ever-changing schema. Network architectures and manual deployment methods from the last decade aren’t suited for this type of environment because they impede enterprise adoption of production deployments at scale.
- Significant Performance Delays: Using containers at scale with tools traditionally used only by DevOps teams creates roadblocks and performance delays for network operators.
A common networking toolset for containers
Cumulus Host Pack aims to enable a path to web-scale networking that supports containers and microservices by utilizing a common toolset of the Linux ecosystem. Because Cumulus Linux brings the host to the network, Host Pack gives application developers and network operators visibility and connectivity of the network. Key capabilities and benefits of Host Pack include:
- Granular container visibility for faster debugging: Host Pack gives operational and development teams shared visibility of application availability through container orchestration tools such as Mesosphere, Kubernetes, and Docker Swarm. Enabled by NetQ running on the host, network operators can view the health of container services, keep track of container locations, track IP addresses and open ports, and have insights into where an issue resides, allowing for faster troubleshooting.
- Simplified network connectivity for improved performance: With the use of routing protocols such as FRRouting and BGP unnumbered directly on the host and in a Layer 3 architecture, Cumulus’ network fabric is able to dynamically learn about containers and distribute these addresses throughout the network to ensure predictable performance between containers across host environments. This removes the complications of a Layer 2 overhead, provides rich and reliable multipathing, simplifies IP address management, and increases reliability. Â
- A common data center operating model, Linux, from network to containers: Cumulus Linux utilizes the same Linux networking model that is foundational to container systems. This enables the use of a common operational toolset, guarantees interoperability, and reduces complexity across the entire data center.
MyPOV
I’ve long predicted a gradual demise of traditional networking vendors. But like other analogous situations, it takes a forcing function to give operators the impetus to change how they work. The appeal that containers offer developers (efficiency, flexibility, modulatory etc.) means that these organizations are looking to containers as potentially the default way of building their new applications. But to do so, they need to resolve the networking and storage problems that this change brings. The timing, therefore, is right for this launch, and Cumulus is likely to ride an ever-growing wave of excitement and adoption of containers. While there are other companies trying to solve the issues around container networking (WeaveWorks being a good example), Cumulus is able to leverage high-credibility founders, investors and, most importantly, reference customers, to gain momentum.