Wednesday, February 16, 2011

OTV Deep Dive - Part One

I've been meaning to do this for a long time and now that I have the blog and am awake in the hotel room at 3AM, what better thing to do than talk about a technology I've been fortunate enough to work with for almost a year. This will be a series of posts as I'd like to take a structured approach to the technology and dig into the details and mechanics as well as operational aspects of the technology.

Overlay Transport Virtualization (OTV) is a feature available on the Nexus 7000 series switches that enables extension of VLANs across Layer 3 networks. This enables new options of data center scale and design that have not been available in the past. The two common use cases I've worked with customers to implement include data center migration and workload mobility. Interestingly, many jump to a multiple physical data center scenario and start to consider stretched clusters and worry about data sync issues and while OTV can provide value in those scenarios it also is a valid solution inside the data center where L3 interconnects may segment the network but the need for mobility is present.

OTV is significant in its ability to provide this extension without the hassles and challenges associated with traditional Layer 2 extension such as merging STP domains, MAC learning and flooding. OTV is designed to drop STP BPDUs across the Overlay interface which means STP domains on each side of the L3 network are not merged. This is significant in that it minimizes fate sharing where a STP event in one domain ripples to other domains. Additionally OTV uses IS-IS at its control plane to advertise MAC addresses and provide capabilities such as loop avoidance and optimized traffic handling. Finally, OTV doesn't have state that needs maintained as is required with pseudo wire transports like EoMPLS and VPLS. OTV is an encapsulating technology and as such add a 42 byte header to each frame transported across the Overlay. Below is the frame format in more detail.

We'll start defining the components and interfaces used when discussing OTV. Refer the topology below.

We have a typical data center aggregation layer based on Nexus 7000 which is our boundary between Layer 2 and Layer 3. The two switches, Agg1 and Agg2 utilize a Nexus technology, virtual Port Channel (vPC) to provide multi-chassis Etherchannel (MCEC) to the OTV Edge devices. In this topology, the OTV edge devices happen to be Virtual Device Contexts (VDC) that share the same sheet metal as the Agg switches but are logically separate. We'll dig into VDCs more in future blog posts, but know that VDCs are a very, very powerful feature within NX-OS on the Nexus 7000.

Three primary interfaces are used in OTV. The internal interface as its name implies is internal to OTV and is where the VLANs that are to be extended are brought in to the OTV network. These are normal Ethernet interfaces running at Layer 2 and can be trunks or access ports depending on your network's needs. It is important to note that the internal interfaces *DO* participate in STP and as such, considerations such as rootguard and appropriate STP prioritization should be taken into account. In most topologies you wouldn't want, or need the OTV edge device to be the root though if that works in your topology, OTV will work as desired.

The next interface is the join interface which is where the encapsulated L2 frames are placed on the L3 network for transport to the appropriate OTV edge device. The join interface has an IP address and behaves much as a client in that it issues IGMP requests to join the OTV multicast control group. In some topologies it is desirable to have the join interface participate in a dynamic routing protocol and that is not a problem either. As mentioned earlier, OTV encapsulates traffic and adds a 42 byte header to each packet so it may be prudent to ensure your transit network can support packets larger than 1500 bytes. Though not a requirement, performance may suffer if jumbo frames are not supported.

Finally, the Overlay interface is where OTV specific configuration options are applied to define key attributes such as multicast control groups, VLANs to be extended and join interfaces. The Overlay interface is where the (in)famous 5 commands to enable OTV are entered though anyone who's worked with the technology recognize more than 5 commands are needed for a successful implementation. :) The Overlay interface is similar to a Loopback interface in that it's a virtual interface.

In the next post, we'll discuss the initial OTV configuration and multi-homing capabilities in more detail. As always, I welcome your comments and feedback.


  1. Great write up Ron! Looking forward to the OTV configuration post. I have a question though, what makes OTV different/better than VPLS? I mean from what I have read so far, OTV encapsulates the VLAN information in packets and forwards them to the remote DC. But VPLS provides a pseudo-multiaccess network, which can also trunk at L2 and carry VLANs as needed. It seems that OTV has the upper hand if your edge is L3? Or you're can't run L2 on the edge for some reason?

    Ziyad B. :)

  2. Hi Ziyad,
    There are a number of differences between VLPS and OTV. I'll try to hit the major ones and we'll do more detial in a subsequent post.
    VPLS requires the following:
    MPLS infrastructure
    Commands per device, per VLAN to establish end-to-end connectivity
    Mesh of psuedo-wires can be complex in larger topologies N*(N-1)/2 for full mesh
    Uses traditional MAC flooding for learning

    OTV requires the following
    IP based infrastrucutre - this could ride over MPLS but MPLS isn't required
    Multicast enabled network (today this will change in future)
    Conversational MAC learning and ARP intelligence to optimize the network and minimize flooding

    We'll certainly cover more detail in future posts - thank you for reading!

  3. OTV is uncompared to VXLAN but what to come from OTV?

  4. Thanks .. This is quite good.. I am following your blog now :-)