Re: [Dnsmasq-discuss] Why does the dnsmasq routing feature require a subnet prefix length of 64?
On 21/6/23 09:49, imn...@gmail.com wrote: I did some math a while back. IPv6 will 'never' run out of addresses? Hah! It'll happen sooner than anyone thinks. - Assume 2^31 IPv6 LANs attached to the internet around the world. And what's stopping more than 2^31 LANs being attached? Why did you focus on what's wasted rather than what's unallocated? It's actually worse than you think, as a lot of service providers are issuing /48s to customers, allowing for 2^16 LANs each, while most users will only have 1 or 2. Hamish ___ Dnsmasq-discuss mailing list Dnsmasq-discuss@lists.thekelleys.org.uk https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss
Re: [Dnsmasq-discuss] Why does the dnsmasq routing feature require a subnet prefix length of 64?
Neal: You aren’t the only one who thought the math was off with IPv6. I had my issues, but for different reasons. Interesting read. R Sent from my iPhone > On Jun 20, 2023, at 7:17 PM, imn...@gmail.com wrote: > > I did some math a while back. IPv6 will 'never' run out of addresses? Hah! > It'll happen sooner than anyone thinks. > > - Assume 2^31 IPv6 LANs attached to the internet around the world. > - Compute 2^31 * 2^64 = 2^95 addresses assigned > - Assume 16 devices connected on each LAN: 2^31 * 2^4 = 2^35 addresses in use > > Converting to decminal, about 40 * 10^27 addresses assigned, 34 * 10^9 > addresses used. That leaves about 1.2 quintillion times the number of > addresses in use that will never be used. > > Had they used /96 as the standard size (32-bit host address), that would've > resulted in about 2^63 addresses assigned for the same 2^35 addresses used. > The wastage would've dropped to about 270 million times the addresses used: > about 12 orders of magnitude less address wastage. > > My opinion on this in more detail: http://murent.us/#ipv6wastage. > > I read somewhere that some may be second-guessing that decision. They > might've done better to use /96 and hash the MAC address down to 24 bits to > make SLAAC work. > > Neal > > >> On Tue, 20 Jun 2023 15:05:07 -0700 >> Eric Fahlgren wrote: >> >> Yeah, some of the RFCs on v6 address formats hem and haw about how big the >> network ID and interface ID parts are (probably written before actual >> implementations were in place), but >> https://www.rfc-editor.org/rfc/rfc4291#section-2.5.1 says quite >> unequivocally: >> >> For all unicast addresses, except those that start with the binary >> value 000, Interface IDs are required to be 64 bits long... >> >> Which drives a stake in the ground regarding how to partition those 128 bits. >> >> >>> On Tue, Jun 20, 2023 at 11:59 AM Petr Menšík wrote: >>> >>> I think that is required by SLAAC RFC, which adds another 2 bytes to 6 >>> bytes of hardware ethernet address. >>> >>> Which is in total 8 bytes, therefore 64 bits is required for it. Prefix >>> cannot be higher, but can be lower in theory. There might be some >>> implementation details now supporting lower prefix length in current >>> implementation. >>> >>> Cheers, >>> Petr On 15. 06. 23 12:07, renmingshuai via Dnsmasq-discuss wrote: >>> >>> When ra-only, slaac, or ra-stateless is configured in dhcp-range and the >>> prefix len is set to a value other than 64, like this: >>> >>> “dhcp-range=2000:1000:1000:1000:1000:1000::, ra-stateless,120,infinite” >>> >>> the following error message is displayed: >>> >>> dnsmasq: prefix length must be exactly 64 for RA subnets at line 16 of >>> /etc/dnsmasq.conf >>> >>> Why must the prefix length be 64? This may come from an RFC regulation or >>> recommendation, but I didn't find it. Would you mind tell me the reason? >>> >>> -- >>> Petr Menšík >>> Software Engineer, RHEL >>> Red Hat, http://www.redhat.com/ >>> PGP: DFCF908DB7C87E8E529925BC4931CA5B6C9FC5CB >>> >>> ___ >>> Dnsmasq-discuss mailing list >>> Dnsmasq-discuss@lists.thekelleys.org.uk >>> https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss >>> > > > ___ > Dnsmasq-discuss mailing list > Dnsmasq-discuss@lists.thekelleys.org.uk > https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss ___ Dnsmasq-discuss mailing list Dnsmasq-discuss@lists.thekelleys.org.uk https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss
Re: [Dnsmasq-discuss] Why does the dnsmasq routing feature require a subnet prefix length of 64?
I did some math a while back. IPv6 will 'never' run out of addresses? Hah! It'll happen sooner than anyone thinks. - Assume 2^31 IPv6 LANs attached to the internet around the world. - Compute 2^31 * 2^64 = 2^95 addresses assigned - Assume 16 devices connected on each LAN: 2^31 * 2^4 = 2^35 addresses in use Converting to decminal, about 40 * 10^27 addresses assigned, 34 * 10^9 addresses used. That leaves about 1.2 quintillion times the number of addresses in use that will never be used. Had they used /96 as the standard size (32-bit host address), that would've resulted in about 2^63 addresses assigned for the same 2^35 addresses used. The wastage would've dropped to about 270 million times the addresses used: about 12 orders of magnitude less address wastage. My opinion on this in more detail: http://murent.us/#ipv6wastage. I read somewhere that some may be second-guessing that decision. They might've done better to use /96 and hash the MAC address down to 24 bits to make SLAAC work. Neal On Tue, 20 Jun 2023 15:05:07 -0700 Eric Fahlgren wrote: > Yeah, some of the RFCs on v6 address formats hem and haw about how big the > network ID and interface ID parts are (probably written before actual > implementations were in place), but > https://www.rfc-editor.org/rfc/rfc4291#section-2.5.1 says quite > unequivocally: > >For all unicast addresses, except those that start with the binary >value 000, Interface IDs are required to be 64 bits long... > > Which drives a stake in the ground regarding how to partition those 128 bits. > > > On Tue, Jun 20, 2023 at 11:59 AM Petr Menšík wrote: > > > I think that is required by SLAAC RFC, which adds another 2 bytes to 6 > > bytes of hardware ethernet address. > > > > Which is in total 8 bytes, therefore 64 bits is required for it. Prefix > > cannot be higher, but can be lower in theory. There might be some > > implementation details now supporting lower prefix length in current > > implementation. > > > > Cheers, > > Petr > > On 15. 06. 23 12:07, renmingshuai via Dnsmasq-discuss wrote: > > > > When ra-only, slaac, or ra-stateless is configured in dhcp-range and the > > prefix len is set to a value other than 64, like this: > > > > “dhcp-range=2000:1000:1000:1000:1000:1000::, ra-stateless,120,infinite” > > > > the following error message is displayed: > > > > dnsmasq: prefix length must be exactly 64 for RA subnets at line 16 of > > /etc/dnsmasq.conf > > > > Why must the prefix length be 64? This may come from an RFC regulation or > > recommendation, but I didn't find it. Would you mind tell me the reason? > > > > -- > > Petr Menšík > > Software Engineer, RHEL > > Red Hat, http://www.redhat.com/ > > PGP: DFCF908DB7C87E8E529925BC4931CA5B6C9FC5CB > > > > ___ > > Dnsmasq-discuss mailing list > > Dnsmasq-discuss@lists.thekelleys.org.uk > > https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss > > ___ Dnsmasq-discuss mailing list Dnsmasq-discuss@lists.thekelleys.org.uk https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss
Re: [Dnsmasq-discuss] Why does the dnsmasq routing feature require a subnet prefix length of 64?
Yeah, some of the RFCs on v6 address formats hem and haw about how big the network ID and interface ID parts are (probably written before actual implementations were in place), but https://www.rfc-editor.org/rfc/rfc4291#section-2.5.1 says quite unequivocally: For all unicast addresses, except those that start with the binary value 000, Interface IDs are required to be 64 bits long... Which drives a stake in the ground regarding how to partition those 128 bits. On Tue, Jun 20, 2023 at 11:59 AM Petr Menšík wrote: > I think that is required by SLAAC RFC, which adds another 2 bytes to 6 > bytes of hardware ethernet address. > > Which is in total 8 bytes, therefore 64 bits is required for it. Prefix > cannot be higher, but can be lower in theory. There might be some > implementation details now supporting lower prefix length in current > implementation. > > Cheers, > Petr > On 15. 06. 23 12:07, renmingshuai via Dnsmasq-discuss wrote: > > When ra-only, slaac, or ra-stateless is configured in dhcp-range and the > prefix len is set to a value other than 64, like this: > > “dhcp-range=2000:1000:1000:1000:1000:1000::, ra-stateless,120,infinite” > > the following error message is displayed: > > dnsmasq: prefix length must be exactly 64 for RA subnets at line 16 of > /etc/dnsmasq.conf > > Why must the prefix length be 64? This may come from an RFC regulation or > recommendation, but I didn't find it. Would you mind tell me the reason? > > -- > Petr Menšík > Software Engineer, RHEL > Red Hat, http://www.redhat.com/ > PGP: DFCF908DB7C87E8E529925BC4931CA5B6C9FC5CB > > ___ > Dnsmasq-discuss mailing list > Dnsmasq-discuss@lists.thekelleys.org.uk > https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss > ___ Dnsmasq-discuss mailing list Dnsmasq-discuss@lists.thekelleys.org.uk https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss
Re: [Dnsmasq-discuss] Why does the dnsmasq routing feature require a subnet prefix length of 64?
I think that is required by SLAAC RFC, which adds another 2 bytes to 6 bytes of hardware ethernet address. Which is in total 8 bytes, therefore 64 bits is required for it. Prefix cannot be higher, but can be lower in theory. There might be some implementation details now supporting lower prefix length in current implementation. Cheers, Petr On 15. 06. 23 12:07, renmingshuai via Dnsmasq-discuss wrote: When ra-only, slaac, or ra-stateless is configured in dhcp-range and the prefix len is set to a value other than 64, like this: “dhcp-range=2000:1000:1000:1000:1000:1000::, ra-stateless,120,infinite” the following error message is displayed: dnsmasq: prefix length must be exactly 64 for RA subnets at line 16 of /etc/dnsmasq.conf Why must the prefix length be 64? This may come from an RFC regulation or recommendation, but I didn't find it. Would you mind tell me the reason? -- Petr Menšík Software Engineer, RHEL Red Hat,http://www.redhat.com/ PGP: DFCF908DB7C87E8E529925BC4931CA5B6C9FC5CB ___ Dnsmasq-discuss mailing list Dnsmasq-discuss@lists.thekelleys.org.uk https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss
[Dnsmasq-discuss] Why does the dnsmasq routing feature require a subnet prefix length of 64?
When ra-only, slaac, or ra-stateless is configured in dhcp-range and the prefix len is set to a value other than 64, like this: "dhcp-range=2000:1000:1000:1000:1000:1000::, ra-stateless,120,infinite" the following error message is displayed: dnsmasq: prefix length must be exactly 64 for RA subnets at line 16 of /etc/dnsmasq.conf Why must the prefix length be 64? This may come from an RFC regulation or recommendation, but I didn't find it. Would you mind tell me the reason? ___ Dnsmasq-discuss mailing list Dnsmasq-discuss@lists.thekelleys.org.uk https://lists.thekelleys.org.uk/cgi-bin/mailman/listinfo/dnsmasq-discuss
