How startup Pluribus looks to commoditize hardware via virtualization
Sunay Tripathi, Pluribus cofounder and CTO
Image: Pluribus Networks
Legacy network vendors have not embraced hardware virtualization in the same way that the server industry has embraced hardware abstraction. Large vendors have embraced either network virtualization or network management via application policies.
VMware has embraced network virtualization via overlays. Juniper and Brocade have leveraged open source solutions to add network virtualization and flow control to their ecosystems. Cisco's Application Centric Infrastructure (ACI) is the latest incarnation of its Open Network Environment or the programmable network.
Very few vendors look at virtualization of the physical hardware. One startup that takes this approach is Pluribus Networks, which made news recently because of a $50 million investment round. Pluribus looks to embrace the commoditization of hardware via virtualization.
Type 1 Network Hypervisor
I spoke with the company's cofounder and CTO, Sunay Tripathi, last year regarding their solution. In development for four years at that point, the key differentiator in the Pluribus product was the approach to network device virtualization. Instead of taking an overlay approach that is popular with products such as VMware's NSX, Pluribus virtualizes the actual network silicon.
According to Tripathi, network vendor software development kits (SDKs) are all single-threaded. Leveraging the SDK would limit the performance of the resulting solution's features, such as flow control in multi-tenant use cases. Tripathi's team set out to create driver-level integration with network chipsets. Creating chipset-level drivers allow Pluribus virtualization software to run on bare metal hardware. VMware's vSphere and Microsoft's Hyper-V leverage bare metal hardware for server virtualization.
Pluribus calls its network device virtualization solution Netvisor. Because Pluribus is writing for the chipset directly, its platform can be leveraged to run on Original Design Manufacturer (ODM) equipment. The potential use case means that Netvisor could not only run on white box hardware but hardware from Cisco and Juniper, for example.
Due to the relative lack of general compute power of ODM hardware, Pluribus has primarily positioned Netvisor as a solution that runs on its white-box hardware. Similar to Cumulus Networks, customers can run standard Linux code on Pluribus hardware; this introduces interesting use cases, including running data center management solutions such as OpenStack directly on switching gear.
Is it enterprise ready?
Networking expert Tom Hollingsworth of Gestalt IT spoke to the limitations of the Pluribus platform on the Jan. 29, 2015 episode of the In Tech We Trust podcast. While Hollingsworth is impressed with the virtualization and performance potential of Netvisor, he discussed concern over features that typical enterprise customers require such as distributed gateways.
I contacted the CTO of Pluribus to see if the company had any comment about Hollingsworth's concerns. I have not received a reply at the time of publication. If I do receive a comment from him, I will post it in this article's discussion.
The bottom line
If your enterprise is comfortable with a DevOps culture and looking for a platform that can be leverage internal intellectual property, then Netvisor offers a fresh take on network device virtualization.
