Spencer Dawkins wrote:
On Sep 28, 2005, at 11:28 AM, Iljitsch van Beijnum wrote:But isn't there still the huge deployability problem caused by having to make upper layers and/or apps ignore the upper 64 bits in many places and make the lower 64 bits globally unique?There's a huge deployability problem with getting shim6 into the hosts too.TonyI guess we're saying that huge=host IP implementations < huge=host IP implementations plus transports plus applications, but I take Tony's point that we are talking about relatively huge changes in either case :-)Spencer
The reason I think shim6 is intrinsically much more deployable than 8+8 is that it doesn't force *any* change on ULPs. They can go on using the current socket API for a hundred years if they want to. What has to be deployed is a new IPv6 stack. One host at a time. We know how to do that. It will take years, but it will happen, due to routine operating system upgrades. And there are no discontinuities of service - just progressive growth in the fraction of sessions that can benefit from shim6. Yes, it is a big change to the stack, but it's *only* the stack. (I don't mean there won't also be changes in TE... but they are not required for shim6 to be deployed and to support failover.) Brian