VXLAN with Dell EMC SDN EcoSystem Solutions

In Part 3, we had a look at Dell EMC OS9’s Static VXLAN implementation; there was also a brief discussion of upcoming OS 10’s MP-BGP EVPN offering – I might dedicate a post to it once it becomes available.

There are a number of DCI options within the Dell EMC SDN EcoSystem, which I will review below.

VXLAN with Dell EMC Networking ON/SDN Ecosystem Partnerships.

Cumulus, Big Switch, Pluribus and IPInfusion have their respective offerings around VXLAN.

Pluribus 

Netvisor/ACF VXLAN Solution is Multicast free, and EVPN like (courtesy of vPorts). It scales to 36 VTEPs on 40G platforms. – I am going to cover their DCI solution in detail in a separate post, but some of the highlights include:

• Multi-point Solution: VXLAN based DCI.

  This solution has an incredibly rich feature set, with support for

  • vLAG (Pluribus MLAG),
  • VRRP (Active active, VIP based),
  • ECMP,
  • VTEP High Availability (VXLAN fabric Edge) and
  • Any Cast Gateways (L3 for East-West at first hop)
  • in addition to the above, it uses its virtual, distributed endpoint database (vPort) to achieve optimizations around endpoint discovery, ARP proxy and forwarding among other things.

• Point-to-Point (Pseudo-wire over VXLAN) service:

  Pluribus call this vLE, or Virtual Link Extension. Legacy pseudo-wire usually referred to services like EoMPLS, of Ethernet over MPLS. If the goal is to connect two sites, vLE is a worthy contender. It is transparent, which indeed implies it can carry control traffic (e.g. for LACP too). Consequently, one could bundle vLEs from VTEP pairs on one site to the other, and vice versa, ending with physical/logical redundancy.

IP Infusion

As of OCNoS 1.3.2, support EVPN based VXLAN. However, there is currently no support for routing over VXLAN, or multi-homing – this will however change in the coming releases. It offers Layer 2 (VPLS and VPWS) as well as Layer 3 VPN Services, along with MPLS.

Big Switch

BCF VXLAN solution is multicast based, and controller driven.The controller requirement is per site, and taking HA into account, translates into 2 controllers per site.

• VXLAN with BCF would extend a L2 segment to another BCF pod/VXLAN network. As this L2 segment is extended between the two sites, it would consequently be common to the them, resulting in Layer 2 adjacency between the two.
• As this extended segment will be contained within a Tenant, which will also typically have other Layer 2 segments, Routing (RIOT function) between the extended segment and other segments within the particular tenant will take place through the tenant logical router, which will not require any further configuration.
• While the BCF Controller itself is the VTEP, physically it will require (a pair of) switches which are capable of VTEP termination, in the leaf group executing the function. These switches will host the member interfaces for the respective interface group.
• It scales to 16 VTEPs as of 4.6.0.

Cumulus

offers a rich VXLAN portfolio, with support for EVPN, Static as well as Lightweight Network Virtualization (Controller-less) VXLAN solutions . It can integrate with VMWare NSX for provisioning.

• Static

  recommended only for small scale.

• LNV, Light weight Network Virtualization

  Controller less, runs VXLAN Service/Registration daemons on Cumulus Linux itself. LNV can support both Service Node replication, as well as Head End replication. However, note the following:

  • HER is the ability to replicate BUM traffic in hardware, providing you have a supported chip. thus, BUM replication happens on your Leaf VTEP ToRs, in contrast to the following – SNR. HER supported scale is 128 VTEPs (Tested). HER is the default option.
  • SNR requires the spine switches to replicate packets in software. SNR could be used when you want to scale beyond 128 VTEPs in LNV model, however the fact that packets are being replicated via software/CPU, on the spine, is a penalty you have to seriously consider – this is not recommended.

• EVPN

  This has support for ARP/ND Suppression. Do note that ARP suppression does not fully inter-operate with Active Active VXLAN, however, that does not impact forwarding. BUM replication is exclusively via Head End replication, unlike LNV.

I have tried to capture some comparative data on the respective VXLAN offerings, but please note:

1. This is by no means comprehensive.
2. This will need frequent revision, so the information may not be current by the time you view it.
3. Ergo, execute your own due diligence. Treat this only as an aid/pointer.
4. I have not included a column for Dell EMC OS 10 MP-BGP EVPN, as this is not released just yet. for now, another missing column is for IPInfusion, I might update the following subsequently.

In the next part, i will focus exclusively on Pluribus ONVL/ACF’s VXLAN Solution