> > > It seems like this is a bug in Amber Road, best dealt with there. > > > > False. > > Can you please elaborate. Why is it OK for Amber road to make > assumptions about interface naming?
Given that this is an open list, I cannot go into the details of the Fishworks clustering architecture. However, broadly speaking, the architecture requires that all clustered resources are identical across the heads. This is impossible in the current IB architecture because the partition datalinks are created by the driver itself based on events that are external to the box, and thus there is no way to guarantee that ibd0 on each head will refer to the same pkey. However, while Fishworks is the catalyst, as covered in the case materials, the work stands on its own and brings IB into line with the established administrative model for VLANs/VNICs with Ethernet. That is, IB partitions are modeled identically to Ethernet VLANs. This is the right answer for a large number of reasons (observability, administrative consistency, ability to create partitions when the IB subnet manager is down ...) that have nothing to do with Fishworks. > To be clear, it seems like a particular product is dependent on the > names of interfaces. Yet, customers ought to be able to rename those > interfaces using vanity names. When they create the partition, they choose the name. This works identically to VLANs, VNICs, aggregations, and so forth. They can then rename the link if they want. -- meem