Rumors of Tungsten Fabric’s Death Are Greatly Exaggerated
In an interesting twist of fate, I’ve been called back into the world of OpenStack and more specifically the world of Tungsten Fabric (TF), an open source software-defined networking (SDN) stack that was recently mothballed by the Linux Foundation Networking (LFN).
I’m excited to re-engage with both projects given our long history together and what looks like a bright future for both.
What is TF? It’s a production-grade software-defined networking (SDN) stack that works across multiple cloud native cloud stacks. Originally based on a commercial product from Juniper Networks, called Contrail Networking (and prior to that, the erstwhile open source project, Open Contrail), TF was open sourced in 2017.
The process of which, was led by yours truly. Juniper Networks customers such as AT&T, eBay, NTT, and Workday wanted to see an open source offering of Contrail Networking in order to avoid vendor lock-in. To make this happen, while at Juniper Networks I led the (re) open sourcing of TF, rebranding from Contrail/OpenContrail, and resulting in its move to the Linux Foundation Networking (LFN).
Using standard interfaces such as Neutron (networking component of OpenStack) and Container Network Interface (CNI), TF integrates with all OpenStack, CloudStack, and Kubernetes distributions delivering true multicloud, hybrid SDN orchestration for virtualized switching, routing, security, Network Address Translation (NAT), microsegmentation, load balancing, and more. Its stability and features are part of why it is relied on for telco cloud by British Telecom, Deutsche Telekom, Etisalat, and Saudi Telecom, amongst others.
Unique amongst many SDN solutions, the TF vRouter forwarding plane delivers high-performance networking for VM, container, and bare-metal server workloads with kernel, Data Plane Development Kit (DPDK), and SmartNIC implementations to optimize CPU resources, space, and cost.
Also unique amongst most SDN solutions, TF provides detailed analytics based on Prometheus and Grafana that integrate with existing cloud ecosystem components, giving network operators a centralized platform for robust insight into SDN operations, cluster health, and diagnostics. Optional flow monitoring employs InfluxDB.
So, why am I working with TF and OpenStack again after six years? As one of the original advocates of the OpenStack movement, an inaugural individual director on the OpenStack Foundation board (now the Open Infrastructure Foundation), and one of the folks who built a production-grade product based on OpenStack and Tungsten Fabric (Cloudscaling Open Cloud System), I feel uniquely suited to helping resolve odd situations in open source land. That’s what we have here when the LFN put TF into “archive” status recently.
But was it the right time for this? I decided to determine if the TF community was actually dead or if Juniper walking away from the table and the LFN more generally had just been a premature assessment of the community’s health.
So in early November, I started reaching out and talking to companies using TF or Contrail in production. I asked if they were still using it, if they were making changes in a non-public fork, and if they had developer resources they would be willing to commit if the community was resurrected without Juniper Networks.
Much Interest in Tungsten Fabric
In my research, I found that there was significant interest in keeping it alive. More importantly, development work was alive and well inside of private repositories and many former community members were considering forking TF and continuing its development on their own in isolation. But did that make sense? Why operate in isolation when one can be part of a thriving community?
What do these companies look like? One example is Nipa Cloud in Bangkok, the largest native Thai public cloud in Thailand. Although Nipa Cloud looked at alternatives, they found that only TF could provide the features, stability, and long-term capabilities they were looking for. Dr. Abhisak Chulya, CEO of Nipa Cloud recently told me that in their search for an open source SDN, only TF met their needs for multicloud and hybrid cloud networking in a robust, secure, and stable manner.
Similarly, ShapeBlue, the CloudStack Company, bet large on TF, right before LFN put it in archive status. ShapeBlue, similar to Nipa Cloud, had assessed their open source SDN options and determined that Tungsten Fabric was their best option for production-grade networking in CloudStack.
Giles Sirett, CEO of ShapeBlue remarked to me that they needed a robust open source SDN solution and that after much research and testing, TF emerged as their best option. Interestingly, ShapeBlue is seeing significant traction with customers looking to replace a combination of VMware ESXi + NSX with CloudStack + TF.
As I canvassed the former TF community I successfully identified 10+ companies with developer resources who were already working on TF code behind closed doors and who were simultaneously thrilled with the idea of coming out into the open and reviving the community. I also identified another 10+ companies with a vested interest in TF’s present and future. That’s more than enough for critical community mass.
Amongst the more interesting companies I expect to participate in next week’s kickoff meeting are Nipa Cloud (Thailand), ShapeBlue (UK), Mirantis (Global), Codilime (Europe), Vinadata (Vietnam), and Nebius AI Cloud (Netherlands).
To that end, the TF community held a virtual gathering earlier today, as described in this LinkedIn posting. A recording will be available if you’d like to get up to speed with the current state of the smaller, more committed community.
If you too see the need for a powerful, open source SDN that’s tuned to the needs of massive-scale operators, join us in helping build the future for TF. As you can see from the others interested in this same goal, rumors of its death are greatly exaggerated.