Move over SDN; startup looking to go where only Cisco, VMware tread

Startup Pluribus takes aim at software-defined fabric

Software-defined network, software-defined data center ... and now software-defined fabric.

Startup Pluribus Networks is proposing a top-of-rack server/switch combination that can run applications, provide storage and provision services like DNS, DHCP, load balancing, firewalls, analytics and packet capture. Through the company's operating system and hypervisor software, two or more of these server/switch combinations can create a large, virtual data center fabric with network virtualization and software-based control with a variety of proprietary and standard APIs.

[ THE LONG AND WINDING ROAD: Figuring out the data center fabric maze ]

Most everyone in the industry is focusing on SDNs: software-defined networks, which seek to virtualize network hardware infrastructures to make them more flexible for VM mobility, changing traffic patterns and added features. Cisco, with its Cisco ONE programmability initiative and UCS fabric computing plan; and VMware, with its Nicira network virtualization platform and overall software-defined data center strategy, probably map closest to what Pluribus is attempting -- or vice versa.

Use cases for the Pluribus software-defined fabric are several:

  • a commodity-priced top-of-rack switch;
  • a Cisco UCS-like infrastructure using Pluribus switches and any vendor's server and storage;
  • cloud computing, with multi-tenancy through isolation, SLAs, QoS, traffic shaping and VXLAN support;
  • SDNs manageable and controllable through OpenFlow, OpenStack Quantum, C and Java APIs, and scripting;
  • low-latency, Arista-like financial trading where parts of an application can run within the server/switch to save hops;
  • application extensibility to create a custom appliance -- native Unix applications and virtual machines, PCI slots for hardware expansion, internal storage for packet capture, and a Clos spine/leaf fabric;
  • and telemetry, with packet capture, lawful intercept and analytics, without the need for separate port mirroring devices.

Pluribus was founded in 2010 by CEO Robert Drost and CTO Sunay Tripathi, both formerly with Sun; and Vice President of Engineering Ken Yang, who is on sabbatical from UCLA. The company, based in Palo Alto, has raised $42 million from investors NEA, Mohr Davidow Ventures and Menlo Ventures.

The company saw a need to have the various components in a data center -- multivendor servers, switches, hypervisors, storage arrays, applications and operating systems -- coordinated and controlled consistently. That's the aim behind Pluribus' Netvisor operating system and hypervisor.

Netvisor runs on the Pluribus F64 server/switch. It's a distributed network hypervisor and operating system that controls and manages the Pluribus fabric and allocates pooled resources and services to virtual networks based on policies and privileges.

A Pluribus API in Netvisor allows the operating system to support other switches and chips, and enable them to be programmed, company officials say. It also supports multipath forwarding so traffic within the fabric can run east/west between racks without having to go through the core switching layer, they say.

The virtual networking component of Netvisor is VNET. VNET acts as a physical network, providing isolation, resources, guarantees and other services.

VNET assigns physical resources like CPUs, memory, bandwidth, ports and hardware queues into virtual resource groups (VRG) that separate VNETs van use. It also pools IP addresses that hosts and VMs in a VNET can use.

Isolation in a VNET can be achieved through VLANs, VXLAN and IEEE 802.1Qbg, otherwise known as Ethernet Virtual Bridging. Control of VNETs can be through OpenFlow, OpenStack/Quantum, through the Pluribus API and CLI or GUI, C, Unix, scripts, etc.

The F64 hardware is a commodity, merchant silicon-based white box platform. It's a 2RU unit based on Intel/Fulcrum Alta chips, with "large" memory, Pluribus says, and extra PCIe slots for expansion.

The switch is low latency, the company claims, with 4 x 40G ports and 48 x 1G/10G ports. The F64 also includes 4 x 2.5 customer-accessible drive bays for hard disk and solid state drives, and Fusion-io flash storage.

The F64 server/switch and NetVisor software are in limited availability, with general availability in the second quarter.

Jim Duffy has been covering technology for over 25 years, 21 at Network World. He also writes The Cisco Connection blog and can be reached on Twitter @Jim_Duffy.

Copyright © 2013 IDG Communications, Inc.

The 10 most powerful companies in enterprise networking 2022