DPDK basic configuration questions

[Michael Mason <michael.l.mason@gmail.com>]

Hi,

We’re trying to determine whether DPDK would be a good fit for a few projects we’re working on.  However, I’m still a bit unclear on how DPDK might fit into our architecture.

Ok, here are some random questions:

1) If one wanted to quickly transfer data from machine A to machine B (A -> B) using DPDK, do both machines need to have DPDK installed?  How about only the receiving side?  

When the recv side binds a NIC to the DPDK driver, the NIC becomes invisible to the OS/kernel.  So, how does this affect machine B trying to send packets to Machine A.  Can Machine B send packets to Machine A’s old/previous IP address?  That doesn’t sound like it’d work since the kernel would not be able to answer ARP requests for the DPDK-enabled NIC on Machine A since it doesn’t know about it any more.  Is this where static ARP table entries come into play?

If (for some reason) we can’t use the prior IP address, can we arbitrarily assign IP addrs and then set up the appropriate static ARP table entries?

Let’s say we’re using link bonding, how would this affect the above scenario?  With link bonding are there multiple IPs that can be used to send data to on Machine A?  

How often do people use network cables to directly connect the machines together instead of using a more traditional networking set up (switches, routers, etc)?

Thank you so much for your time!
~mlm