FCoE vFC interface binding

[pdotchon]

Hello,

I am looking for some help/explanation of the requirements for virtual FC interface binding to either an interface or mac address. I am confused about why this is required, and why both options exist. The following relates to a Nexus 5596 configuration.

As I understand it, a vFC must have the 'bind' command in order to associate the vFC interface directly to either:

1. a host CNA directly attached to an Ethernet interface (bind interface), or

2. a host CNA attached to an intermediate switch (bind mac address)- a FIP snooping switch between the FC Forwarder and the CNA.

Does this mean we need a unique/dedicated vFC interface for each CNA?

I am aware of a currently working configuration where the vFC is bound to an Ethernet interface (single link) that connects to a blade chassis switch, which in turn has a number of blades/CNAs attached. I would not expect this to work, as it effectively means multiple FLOGI requests will arrive on the same vFC interface from multiple CNAs. Is this supported and why does it (apparently) work?

The scenario above did initially have feature NPIV enabled, and I assumed this we the reason it worked (and was also the reason for feature NPIV to exist at all in FCoE environments) but it has since been disabled with no adverse effect.

Can someone give me some background info and a clue as to what in the above is correct/incorrect/supported or not etc?

Thanks!

Phil Dotchon

[pdotchon]

I hope someone can help with this?!

[Jeremy Waldrop]

A unique vfc is required for each CNA. vfc can be bound to an Ethernet interface or port-channel interface.

NPIV is not required for FCoE to function. NPIV is needed when multiple FC/WWPNs (initiators) need to login over the same physical port or port channel.

[alones123]

If the server CNA is directly attached to port at 5K or 2232PP, you can bind port.

If your are using Nexus 4K switch, you need bind MAC to vfc, they are trunked from port-channel, can't direct bind 4K port form 5K. 

4K has 14 10G port facing inside servers, so MAX you need create 14 vfcs binding different server for single 4K blade switch.

[pdotchon · Accepted solution]

Thanks for the responses to my questions. I guess I understand that the configuration guides suggest a unique vFC interface is required for each CNA, but I'm wondering why? Also, what happens if there is more than one CNA connecting to a vFC port.

In our case, we have an IBM EN4093 switch, configured for FIP snooping, between the blades and the N5K. There is a single 10G link between the IBM switch and the N5K, and the vFC port on the N5K is bound to this interface. Multiple blades/CNAs are connecting and working over this single vFC interface, so my question is why should we change this to multiple vFC ports bound to MAC addresses, with all of the associated admin required to manage the MAC address configuration on the Nexus?

Although I sort of recognise the FCoE is essentially trying to mimic normal physical FC N to F port connections, what we have working is multiple N ports connecting to a single F port, which in normal FC equates to NPV. As shown in the link, the FCoE NPV feature does result in a similar topology, and note that the FCF switch apparently supports this without additional configuration, so I wonder if that is why our setup works?

http://www.cisco.com/en/US/docs/switches/datacenter/nexus5000/sw/san_switching/521n11/b_5k_SAN_Switching_Config_521N11_chapter_0110.html

(note the diagram is missing in the html page, but it is visible in the pdf)

regards,

-phil