What's so bad about IPv6? [LWN.net]

What's so bad about IPv6?

Posted Jan 26, 2011 16:43 UTC (Wed) by mstefani (subscriber, #31644) [Link]

Should. How well is a different story.

Some devices will support it only "in software"; hitting the CPU on a busy network devices is basically "game over". Other devices support IPv6 only in a limited way, e.g. no VRF support for IPv6. Of course for those type of problems the network vendors are more than happy to sell you new hardware for $$$ with a questionable ROI. But for some features your network might rely upon all you will get back is a "IPv6 support will be added next year"...

In theory, theory and practice are the same. In practice, they are not.

Posted Jan 26, 2011 17:36 UTC (Wed) by khim (subscriber, #9252) [Link]

No, they don't support it. It's easy, really: you claim that IPv6 is simpler to process and so it should provide performance boost. But let's check the facts

up to 60 million packets per second (Mpps) of IPv4 unicast forwarding traffic and up to 30 Mpps of IPv6 unicast forwarding traffic

Oops? Looks like simple and logical way to save half of equipment is to just disable IPv6 - and this is what ISPs are doing... It does not matter if hardware has support for IPv6 or not if it's disabled.

In theory, theory and practice are the same. In practice, they are not.

Posted Jan 26, 2011 19:06 UTC (Wed) by ebiederm (subscriber, #35028) [Link]

Shrug line rate 10gbps software forwarded ipv6 in linux.

In theory, theory and practice are the same. In practice, they are not.

Posted Jan 26, 2011 21:40 UTC (Wed) by bronson (subscriber, #4806) [Link]

Markov generated text seeded with the iptables manpage?

In theory, theory and practice are the same. In practice, they are not.

Posted Jan 27, 2011 13:51 UTC (Thu) by hmh (subscriber, #3838) [Link]

Must be. It lacks everything, from content, to truth.

Being someone who has actually tried to do 10Gbps routing using Linux, I am well aware of its limitations. You need lots of tuning and the correct hardware to get high packets-per-second rates, and it gets nowhere close to the target 40Mpps. It really is useful only for large packets, or if you need nowhere near line-rate and don't care about DoS attacks with small packets.

One really needs hardware-assisted packet forwarding to do line-speed 10-gigabit routing at all packet sizes. Either that or a routing cluster, at which point TCO goes well above a proper 10Gbit Cisco/Juniper switch-router.

So, the question becomes: are there affordable, non-experimental hardware packet forwarding devices (preferably PCIe) that are compatible with Linux?

What's so bad about IPv6?

Posted Jan 26, 2011 20:08 UTC (Wed) by hmh (subscriber, #3838) [Link]

That is a blatant lie. There are many vendors that still have IPv6 in *roadmap* for several of their product families. I should know, we have several here and it is being an effective deterrent to switching to IPv6.

What's so bad about IPv6?

Posted Jan 26, 2011 18:00 UTC (Wed) by tialaramex (subscriber, #21167) [Link]

Back in 2003 I was testing Cisco's hardware (as much as the IPv4 version is) IPv6 multicast snooping and routing. We were playing different full-rate video streams over a large multi-site IP network with maybe a dozen switch clusters. Without multicast, this was a quagmire (each person viewing incrementally increases the bandwidth needed at pinch points, until performance becomes excruciating) and without IPv6 it requires lots of management because the (Class D in this case) addresses were scarce and must be handled very carefully. With both it was a joy.

Granted, we found several nasty bugs and spent long evenings on transatlantic telephone calls getting their engineers to reproduce them. But that was now almost EIGHT YEARS AGO. This stuff was available for production users when ISPs were buying the gear that's now dusty and old and being thrown out. But they didn't buy it, because less capable gear is always cheaper.

In some way it's our fault as consumers. Most ISPs live on razor thin margins. If everybody was happily paying 50% more for Internet access there'd be money to buy anything the ISP's engineers could want, have real coffee in the DC quiet area, and still buy the Chairman a yacht. But it's a cut-throat business, only specialist players can afford to charge more and deliver better service.

What's so bad about IPv6?

Posted Jan 26, 2011 19:54 UTC (Wed) by cmccabe (guest, #60281) [Link]

> In some way it's our fault as consumers. Most ISPs live on razor thin
> margins. If everybody was happily paying 50% more for Internet access
> there'd be money to buy anything the ISP's engineers could want, have real
> coffee in the DC quiet area, and still buy the Chairman a yacht. But it's
> a cut-throat business, only specialist players can afford to charge more
> and deliver better service.

Our fault as consumers? That's outrageous.

Around here, people routinely get charged $50 a month by Comcast for mediocre internet service. And if Comcast has a technical problem, that requires a technician to solve, they will charge you for the time!

The real problem here is that the interests of the ISPs are not aligned with those of consumers. Consumer interests are best served by open markets and open standards. ISP interests are best served by creating high barriers to entry, locking in consumers, and offering the minimum acceptable level of service.

Nearly every place in the US has either a local ISP monopoly or duopoly. Comcast or AT&T are your choices. They're not even bothering to pretend that the different parts of AT&T are independent any more. It's not a free market, it's a monopoly. And they know how to milk it.