Darren Reed wrote:
My view of IPMP is that it is functionality that is layered on top of NICs - it is a virtual NIC that has the backing of real ones behind it. It isn't doing any real I/O itself, it is passing off packets to real NICs for sending and similarly, packets received via an IPMP interface come via a real NIC first. Maybe this mental picture is a bit premature, given that no detailed or implementation picture has been presented. It is also built on how you can build software NICs on BSD: just write some skeleton code to do something special with packets passed on to other NICs "behind" the software one for further IO. While I don't expect us to necessarily adopt the BSD model, I can't see why we shouldn't make it easier to write software-only NICs. e.g. the faith NIC in NetBSD:
You can already do that in Solaris today (albeit not with "stable" interfaces), aggr exposes NICs that correspond to aggregations of hardware NICs. VNIC which we're working on as part of Crossbow allows multiple virtual NICs to be defined on top of a single hardware NIC. This is all done using the interfaces currently provided by the Nemo framework.
-- Nicolas Droux, Solaris Kernel Networking Sun Microsystems, Inc. http://blogs.sun.com/droux _______________________________________________ networking-discuss mailing list [email protected]
