The vector unit must be FMA capable just like Larrabee, hence the doubling of FLOPS/cycle.

Mpath-tools is a set of programs for linux 2.6+ that aim to facilitate load balancing and failover over multiple and heterogeneous ISP connections.

4.0 was NT4 ... my guess is 95/OSR2/98 were all 3.2x

It was too computationally intensive for *realtime* rendering in 1973, but clearly not out of reach for the kind of modeling software NASA people were using ...

Also, it should be noted that realtime phong shading was already common in demos/intros running on 33 MHz 386 CPUs back in the 90s

IPv4 multi-homing can't be done without BGP, either. The requirements for Provider Independent address space in IPv6 are identical to the requirements for PI address space in IPv4.

I meant "cheap" multi-homing without a PI address block, like it's used in many small/medium offices where you have multiple ISP links and just failover by changing the SNAT mapping at the gateway when one goes down.

Doing this with ipv6 requires renumbering the whole internal network each time you switch links, or the more costly alternative of getting your own PI block, which in turn isn't IMHO sustainable for the long term because everyone doing it would make the global BGP table grow an awful lot faster that it's already.

OK, so it seems you agree (with the Founders of the Internet) that end-to-end is a good thing. .... and now you say it's a bad thing. So is it a good thing or a bad thing then?

My personal opinion is that end-to-end is *generally* a good thing, but shouldn't be *enforced* because there always will be edge cases where it will conflict with privacy.

Er, what??

Do you mean to say that without NAT a firewall is not needed, or that a firewall doesn't impact reachability ?

With IPv6 I can use NAT if I want.

I'm all for freedom of choice, my problem actually is that you can't use ipv6 NAT even if you want. Not with Linux anyway.

If people want anonymity within their local network, then there will be a market for devices that do IPv6 address cloaking and you can buy one and use it to hide your addresses.

Exactly, you would have to pay for something you can achieve with one iptables command line on ipv4. See my point ?

The lack of SNAT/DNAT targets in Linux ip6tables makes it quite impossible to use ipv6 for any serious enterprise networking. Ipv6 multihoming can't be done without BGP, other solutions like mobile ipv6 or shim6 are - at best - a big mess, also who wants to broadcast his internal network topology/numbering scheme to the whole internet ?

There seems to be some kind of religious taboo here, where the only - supposedly - evil use of NAT (N-to-1 mapping) being taken into consideration, but this is IMHO just plain wrong. Also the NAT haters main argument is that it doesn't preserve end to end reachability (which is not even true for N-to-N mappings), but without NAT everyone is gonna use a stateful firewall for ipv6, and guess what ... the effect on reachability is almost exactly the same.

The other problem I have is with anonymity, without NAT every PC in your local network may be identified individually, there are many cases where this may not be desirable.

IMO ipv6 brings some nice new stuff to the table, the most obvious being the xxl address space, but takes away too much for me to consider using it for myself or my customers at the moment.

Another world just stored vector graphics instead of bitmaps, takes less space, makes it much easier to animate too.