Vint Cerf On Broadband, Wireless, IPV6 And More

Carnage4Life writes: "There's a very interesting interview on Upside with Vint Cerf ^[?] who is currently senior vice president for Internet architecture and technology at MCI Worldcom. In the article he discusses the problems facing the current specifications for wireless protocols, UUNet and how it will be adapted to face the future (maybe by becoming an optically switched network), his home wireless network, IPv6 ^[?] and his expectations of how broadband will change the Net. " Ya ever think what the world owes these guys? Wow.

Comments

IPv6

[Signal 11]

I think it'll be some time yet before IPv6 is adopted on the 'net at large. While the infrastructure is getting close to being there, the sad truth is that the "dot coms" will do the typical corporate dance and ignore the problem until it affect their bottom line. "Why upgrade to IPv4? It costs more, and doesn't get us any new customers" they'll say. And then some fine afternoon ARIN will announce we've run out of IP addresses. Then everyone will panic and try to upgrade their hardware overnight.. millions of ISPs will experience network blackouts and brownouts, router loops, and all kinds of other madness as software and hardware updates go awry. Usenetters will proclaim it's the end of the internet.. again. Slashdot will run an article on it.. a few weeks later after billions are wasted in capital everything will start working again, people will forget all about it.. and it'll be blamed on somebody else. Probably us, of course, for not telling them this would happen sooner. As if.

It's mainly about address space...

[Cato]

First of all, I agree IPv6 has some nice features beyond address space - however, they are not all unique to IPv6, many are also in IPv4.

> Speed (simpler headers and simpler routing)

This should happen, but all the ASICs in existing routers need to be changed (unless the designers were very far-sighted and wasted some silicon on v6 support). Headers are bigger in v6, but it should be easier to do fast silicon because they are more regular (header options are faster to process too). As it happens, Intel's IXA and similar network processors may make it easy to do very fast routers using many parallel IXA-like chips (one vendor is doing a 180-IXA-chip router), for which IPv6 is simply a software upgrade.

IPv6 does make it easier to keep the size of core Internet routing tables down, but it seems that stub networks (e.g. enterprises or mobile phone networks) will migrate to v6 first, leaving core networks till last.

Probably speed will not be a deciding factor either way.

> Mobility (mobile IPv4 has relied on all stations involved having forwarding systems)

IPv6 has some advantages for mobility, because mobile IP was built into IPv6 hosts, but I can't remember what they are...

> Autoconfiguration (no more messing with DHCP or BOOTP configuration files)

Yes, but DHCP does more than ND (Neighbour Discovery) and RADV (Router Advertisments) - e.g. it configures DNS servers, domain names, etc. So DHCPv6 will still be needed in many environments.

> Security (IPSec is mandatory)

IPSec is mandatory, but it won't be turned on as default until someone solves the scalability and performance issues of IKE (Internet Key Exchange protocol, which authenticates both parties and sets up keying material). PKI is the only scalable way to do IKE currently, and PKI is a nightmare. Also, IKE has quite long delays (in the seconds) when setting up sessions, which is perhaps why it is typically used between IPSec gateways in tunnel mode.

> Optimised Connections (anycasting allows you to locate the nearest active server of the type you want)

Anycast is very cool, but not yet implemented in the IPv6 stacks I've seen (e.g. Linux). I think the IETF is still working on how this will be implemented.

> Quality of Service (another mandatory feature)

In what sense is QoS mandatory? I have Linux IPv6 set up at home, but I don't have RSVP installed. I work for a company that does QoS provisioning software, and the only QoS feature I can see in IPv6 that is different to IPv4 is the Flow Label (a 16 bit field that optimises the classification of app to app traffic flows, for use with RSVP). The Traffic Class field in IPv6 is identical in format to the TOS byte used in IPv4, and will use DiffServ in the same way.

> Multicasting (yet another mandatory feature)

Not sure exactly which bits are mandatory here, either. Multicast has been designed in, and is probably better supported in IPv6, though I've not looked at this in detail. Multicast routing protocols are a separate issue to IPv6 vs v4, they simply need updating to be able to route IPv6 multicast traffic. There is quite a lot of practical work in network management of multicast to be done still, whether on IPv6 or IPv4, though it is seeing some deployment. QoS is probably a pre-requisite for most people to deploy multicast - until you can control multicast apps' use of your network bandwidth, it's tough to allow them to be deployed except if you control the app servers very tightly.

I think the killer apps for IPv6 are:

* Address Space - this will drag people kicking and screaming into IPv6, in order to support always-on (good point from Vint Cerf about increased duration of IP address usage), lack of massively scalable NATs (let's see someone NAT 10 million cable TV users...), cellphones/smartphones with IP, home appliances, etc.

* Getting rid of NAT hassles - trying to get applications to work through NAT is a pain and sometimes impossible by design (e.g. IPSec transport mode). This is probably not a killer reason, but will help the decision, particularly where the end host must act as a server (e.g. sending short messages or news updates to a mobile phone).

* Mergers and Automatic Network Renumbering - if two companies merge, you currently have to NAT traffic between their networks, or go through the pain of manual renumbering. IPv6 lets you auto-renumber from a single point, everything 'just works'. Since 'within the firewall' applications in a merged network would still have to cross the NAT, and many protocols such as DCOM, CORBA and so on are NAT-hostile, this may be a strong motivation.

Ultimately, address space is the single biggest reason, particularly in Asia (which was late to the Internet and got a tiny allocation, allegedly smaller than some US companies have).

Why so hard to mix IPv4 and v6?

[Noer]

Ok, IPv4 uses 32-bit numbers (four dotted-bytes). IPv6 uses 16 dotted bytes (128 bits). I don't see why the current IPv4 network can't be treated as one network within IPv6, with 12 of the bytes set to a constant. That would make translation pretty easy, though of course software still has to be updated. I.e. 128.45.3.25 for example would map to 1.1.1.1.1.1.1.1.1.1.1.1.128.45.3.25, and the range that's all ones would gradually expand to include other networks.

Re:Why so hard to mix IPv4 and v6?

[I0ta]

Here's the direct quote from RFC2133.txt: "
This address format allows the IPv4 address of an IPv4 node to be represented as an IPv6 address. The IPv4 address is encoded into the low-order 32 bits ofthe IPv6 address, and the high-order 96 bits hold the fixed prefix
0:0:0:0:0:FFFF.
IPv4-mapped addresses are written as follows:
::FFFF:IPv4-address
"
-I0ta

IPV6 transition

[Nicolas MONNET]

IPV6 has been out for years. 5 years? 6 years? And we still haven't got a transition strategy. I mean, the questions Cerf raises in this article are the same that 5 years ago! They are the same than in the IPNG RFC! The truth is: nobody has any idea how to do the transition.

Because, simply, the problem is NOT in infrastructure. Putting IPV6 in the backbones is almost trivial -- I mean, it could be done now already, you just encapsulate IPV4 in some way.

Now ... on the client side, it's another story. There is NOTHING ready on the client side. Absofuckinglutely nothing ready. Oh yeah, a whois client, and a name daemon. Maybe a telnet and FTP. And that's it.

Now, I have to ask myself, as a programmer, how would I do to support IPV6 in my programs? I don't have the slightest idea. I would'nt even be able to test them properly. Would there be an IPV6 compatible Apache, I would'nt be able to use a whole bunch of Perl modules with it. Of course, I would need an IPV6 enabled Perl. Etc, etc ...

This is a BIG problem. A lot of cash has to be thrown into this, like in a consortium or something .. but who will have the incentive to do this?

geeks of the century award

[stickytar]

For what it's worth, Vinton Cerf and Robert Kahn came here to Montana a while back and in a small packed conference room at the university we were able to field some questions to these guys.

I was intrigued by what they had to say about the history of them developing TCP. The whole ipv4 that we have now comes from their original general assumption that at most only 9 network nodes at most (colleges, and research centers) would ever be using this silly thing. Now the much needed? shove towards ipv6 has even our toaster beeming with glee. What impressed me the most wasn't Vinton Cerf, but Robert Kahn. Cerf took the money and ran so to say. He has a BIG job at MCIWorldcom and is highly recoginized, but the real meat and potatoes programmer, Robert Kahn, does not have the big glitz job. He is quietly doing research as we code away at nights. This is the man that deserves recognition for the internet. He did the majority of the coding for the stack and was the driving force at it's implementation. Not to down on Cerf because they are both brilliant minds, but it seems the real coders out their never get their due share. For what it's worth...

-------------------------------------------------- -
refrig: copy that toaster:2 transmitting butter now.

Ooh yeah...

[Sharkey [BAMF]]

"You've got your fiber layer, and you're going to carry some number of wavelengths on each fiber -- the term that is used is lambda for wavelength, so lambda means color, really. One fiber can carry a number of colors -- there could be as many as a hundred, maybe even more. Each color might be transporting as much as a terabit of capacity"I'm sorry, but did that give anyone else an erection, or is it just me?Sharkey
www.badassmofo.com

IPv6 is more than addressspace

[jd]

One of the reasons it ISN'T more widely deployed is that that is all it is perceived as being, by the PHB's with all the cash.

Usually, though, NAT and firewalling give you essentially the address space you want, with no extra deployment costs. Hence, the total lack of interest.

BUT, IPv6 also offers:

• Speed (simpler headers and simpler routing)
• Mobility (mobile IPv4 has relied on all stations involved having forwarding systems)
• Autoconfiguration (no more messing with DHCP or BOOTP configuration files)
• Security (IPSec is mandatory)
• Optimised Connections (anycasting allows you to locate the nearest active server of the type you want)
• Quality of Service (another mandatory feature)
• Multicasting (yet another mandatory feature)

I dunno about you, but I think ISPs that can get a feature list like that would be far more interested than if they're told they get more IP space to sell. There are only a finite number of customers in an area.