Thank you to everyone who joined SDxCentral for its latest DemoFriday featuring Pluribus Networks on SDN switching. Attendees were able to learn how to enjoy the full benefits of software-defined networking (SDN) without being a hyperscale or webscale shop. Netvisor took centerstage and Pluribus Networks CMO Dave Ginsburg demonstrated how it allows for overlay-free virtualization and interoperability with existing networks. Following the presentation on SDN switching, Dave Ginsburg kindly responded to questions from the audience.
Be sure to take a look at the other helpful resources like the full presentation and demo-only portion of the event on the Pluribus DemoFriday archive page.
Can you elaborate on the hardware offload you mentioned?
Pluribus: The Pluribus server–switch architecture creates a high-speed control path between the switch silicon and the control processor, up to 500x that of traditional TOR switches. This permits the CPU memory, up to one terabyte depending upon the platform, to act as an adjunct to the more limited hardware MAC table. The software MAC table can scale to millions of entries.
How were the op-ex savings calculated?
Pluribus: We looked at traditional box-by-box management and then compared it to efficiencies by managing a fabric from a single point. For analytics, we the looked at the added advantage by eliminating the op-ex of a separate visibility fabric. Details are on our featured article on SDDC deployment.
In large Layer 2 networks, are you using STP to break loops, as you don’t use any overlay?
Pluribus: In a Pluribus-only network the software vport table can be used to detect and break loops without STP. Netvisor does support STP for interoperability with non-Pluribus switches.
What application use cases are offered and/or deployed via REST API?
Pluribus: All Netvisor commands are available via the CLI, the REST API, the Java API, and the C API. Future support will include the SAI. Almost all are also available via the vManage GUI.
You mentioned OpenStack integration, could you explain a bit more on level of integration of Netvisor with OpenStack?
Pluribus: Where an external controller is deployed (i.e., RedHat RHEL OSP), the Pluribus node supports the Neutron ML2 plug-in for networking. It also supports multi-tenancy based on Netvisor VNETs, as well as virtualized appliances via KVM. These can be assigned to individual tenants. Netvisor also ships with an integrated Centos OpenStack controller, eliminating the need for a separate server.
How do you detect microbursts in the network caused by congestion/misconfiguration/DDoS?
Pluribus: Netvisor records statistics for ports, VLANs, flows, and TCP connections in rolling log files that can record days, weeks, or months worth of data depending on traffic and how much disk space is available for logging. A microburst in traffic typically shows up in port statistics. These port statistics can then be correlated with VLAN, flow, and transmission control protocol (TCP) connection statistics to identify the reason for the congestion and suggest remediation steps.
Do you support Layer 2 ECMP?
Pluribus: Netvisor supports ECMP routing as well as Active/Active VLAG in clusters.
In addition to CLI & Rest, what traditional northbound interfaces are supported from a vFabric perspective?
How is a Layer 2 link generated logically between two different ports between two different vendor nodes?
Pluribus: Ports can be aggregated manually or using LACP.
Does the additional CPU/memory on the server-switch allow for full BGP routes (as opposed to traditional switches that have a limit typically far less than full routes)?
Pluribus: In theory, yes, but this is not yet implemented in production code.
Can you talk about how NetVisor handles VM migration/movement and lifecycle management? Is your Switch-Server architecture transparent to VM lifecycle issues?
Pluribus: From the application standpoint, it is transparent, but given the embedded analytics, we can track the movement of VMs and set rules (vFlows) accordingly. In the future, we will have closer integration between Netvisor and vSphere.
Are edge router services like MPLS and VPLS supported?
Pluribus: The system is currently positioned as a converged edge platform, but not a classical edge router. Some of the solutions provided by MPLS and VPLS may be solved by the system’s L2/L3/VXLAN support.
Does nvOS support sub-second link protection using BFD over vLAG groups and over VxLAN tunnels?
Pluribus: Netvisor monitors links for failure at the physical layer now. A future release will support BFD.