 RARPSL
join:1999-12-08
Suffern, NY | reply to ncherry
Re: 2^128 Addresses Claim is inaccurate/misleading said by ncherry:said by RARPSL:Check out » en.wikipedia.org/wiki/IPv6_addre···ss_Space. The low 64 bits are the Host Address (per the IPv6 addressing rules). If the ISP follows the IANA guidelines then each customer gets a /48 block (see the last 2 paragraphs in that section). This allows 16 bits of the address for sub-networks on the users LAN. ARGH! I think my head is going to explode!  Thanks for the correction! You are very correct, I'm not sure what to make of this but my gut feeling is that we won't see a /48 prefix length in the home. But I will update my IPV6 page further to reflect the information above (RIR - /23 -> ISP - /32 -> Customer - /48 -> Host /64, addressed using EUI-64). On the rest of your points, Ah! that make much more sense now, sorry I was very confused. IPv6 addressing is going to be very interesting as it is even suggested that we waste a /64 on a point-to-point link (though it does say you could use a /126 or, for the gluttons among us, a /127). Which kind of brings us back to your initial point of the vast expanse of IPv6 addressing being whittled into many heaps of wasted address space. The engineer in me shrieks at the thought of so much wasted address space. It's the same thinking behind '640K ought to be enough for anybody' we see in practice today. You're welcome. What is the address of your IPv6 information page?
Due to the reservation of the low 64 bits for the host address, the longest prefix that can be assigned to the user is a /64 (which gives a single network to the user). A /56, as I noted, gives the same 256 networks as at present with a single IPv4 address backed by a NAT Class-C router (ie: 192.168.x.y/24 addressing) where you have 256 sub-networks on your LAN. If the ISP wants to slice up the /48 block they are supposed to be giving the user, this should be, IMO, a good point to make the boundary (They get 4096 users per /48 block and the user gets the same 256 Sub-Nets as they get with the Home NAT Router). |
|
 ncherryPremium
join:2003-07-13
Monroe Township, NJ | said by RARPSL:You're welcome. What is the address of your IPv6 information page? My home IPV6 notes are at: »www.linuxha.com/other/home-ipv6.html Take it with a grain of salt as I need to clean it up after the last few weeks of information have me rethinking everything.
said by RARPSL:Due to the reservation of the low 64 bits for the host address, the longest prefix that can be assigned to the user is a /64 (which gives a single network to the user). A /56, as I noted, gives the same 256 networks as at present with a single IPv4 address backed by a NAT Class-C router (ie: 192.168.x.y/24 addressing) where you have 256 sub-networks on your LAN. If the ISP wants to slice up the /48 block they are supposed to be giving the user, this should be, IMO, a good point to make the boundary (They get 4096 users per /48 block and the user gets the same 256 Sub-Nets as they get with the Home NAT Router). The /64 seems to be heavily suggested but many of the RFCs speak of larger prefix lengths so I'm certain it's not written in stone. In fact that's the one thing that has me the most confused. It seems it would be a bad idea not to follow the /64 ARIN recommendation as something may break. I still have a lot of reading to do on this.
My engineering concerns are from a consumer prospective (for my home) and from a business/managed network prospective. For the home it has to be as simple as possible. For the business it must be resource conservative (balance costs with performance). I don't think I'd like to address router loopbacks with a /64 (actually a /128 shouldn't break anything) and I'm not happy about wasting a /64 for a point-to-point line (a /127 breaks DAD and it's suggested that it breaks IPv6 Mobility, I could settle for a /126). I should be able to properly summarize the business network to keep the overall route table small.
The concerns of the ISP are somewhat beyond me as I don't do that kind of work. IMO, a huge guess really, is that the 4096 networks are not going to be enough for someone as large as Comcast. Either way, we'll see as Comcast said they'll share what they find and the info will leak out anyway.
As I'm fond of parodying: "My God! It's full of TLAs!" (Three Letter acronyms and paraphrased from 2001 )
--
Neil Cherry
Linux Home Automation
Linux HA Blog
Author: Linux Smart Homes For Dummies
|
|
RARPSL
join:1999-12-08
Suffern, NY | said by ncherry:said by RARPSL:You're welcome. What is the address of your IPv6 information page? My home IPV6 notes are at: » www.linuxha.com/other/home-ipv6.html Take it with a grain of salt as I need to clean it up after the last few weeks of information have me rethinking everything. said by RARPSL:Due to the reservation of the low 64 bits for the host address, the longest prefix that can be assigned to the user is a /64 (which gives a single network to the user). A /56, as I noted, gives the same 256 networks as at present with a single IPv4 address backed by a NAT Class-C router (ie: 192.168.x.y/24 addressing) where you have 256 sub-networks on your LAN. If the ISP wants to slice up the /48 block they are supposed to be giving the user, this should be, IMO, a good point to make the boundary (They get 4096 users per /48 block and the user gets the same 256 Sub-Nets as they get with the Home NAT Router). The /64 seems to be heavily suggested but many of the RFCs speak of larger prefix lengths so I'm certain it's not written in stone. In fact that's the one thing that has me the most confused. It seems it would be a bad idea not to follow the /64 ARIN recommendation as something may break. I still have a lot of reading to do on this. My engineering concerns are from a consumer prospective (for my home) and from a business/managed network prospective. For the home it has to be as simple as possible. For the business it must be resource conservative (balance costs with performance). I don't think I'd like to address router loopbacks with a /64 (actually a /128 shouldn't break anything) and I'm not happy about wasting a /64 for a point-to-point line (a /127 breaks DAD and it's suggested that it breaks IPv6 Mobility, I could settle for a /126). I should be able to properly summarize the business network to keep the overall route table small. The concerns of the ISP are somewhat beyond me as I don't do that kind of work. IMO, a huge guess really, is that the 4096 networks are not going to be enough for someone as large as Comcast. Either way, we'll see as Comcast said they'll share what they find and the info will leak out anyway. As I'm fond of parodying: " My God! It's full of TLAs!" (Three Letter acronyms and paraphrased from 2001 ) The standard code in all the operating systems reserve the low order 64 bits for the host address. Thus they assume (require?) that the user's prefix is no longer than a /64. My comment about not giving everyone a /48 but supplying a longer prefix allows the ISP (Comcast) to support more users by dividing each /48 into a number of separate user prefixes.
The loop back address (equivelent to 127.0.0.1) is ::1 [which is 127 0's and a 1] a reserved address - see the Wikipedia address. Also, your point-to-point uses only two of your assigned addresses since each device will have a host address. If on separate networks (ie: Different prefixes) your end acts as a router gateway. Also, remember that a NIC/Port can have multiple addresses.
The 4096 networks is the number of users per /48 block if they assign /56s to each user. The ISP is assigned a /32 or shorter prefix (/28 for very large ISPs). By assigning a /56 in lieu of a /48 the number of blocks is increased 4096 times. |
|
ncherryPremium
join:2003-07-13
Monroe Township, NJ | said by RARPSL  ...
The standard code in all the operating systems reserve the low order 64 bits for the host address. Thus they assume (require?) that the user's prefix is no longer than a /64. My comment about not giving everyone a /48 but supplying a longer prefix allows the ISP (Comcast) to support more users by dividing each /48 into a number of separate user prefixes.
The loop back address (equivelent to 127.0.0.1) is ::1 [which is 127 0's and a 1 : a reserved address - see the Wikipedia address. Also, your point-to-point uses only two of your assigned addresses since each device will have a host address. If on separate networks (ie: Different prefixes) your end acts as a router gateway. Also, remember that a NIC/Port can have multiple addresses. I understand your point on the /64, neighbor discovery, auto-configuration, mobility etc. all pretty much depend on it. Every host will support it.
Sorry, the "loopbacks" I'm referring to are on the router (specifically Cisco) and are really not a loopback interface (i.e. associated with 127.0.0.1 or ::1). They're more of a virtual LAN (but not a VLAN) associated with no physical interface but useful for things like management and tunnel termination for routers that might have multiple interface. Generally you can set these interfaces to a /128 (IPv6) or /32 (IPv4).
Okay, the next comments are "my general practice" type comments. Engineering always depends on the constraints. Understand that my general work is done on the WAN portion of networks and that different practices apply to the LAN (and even the internet).
On the point-to-point (things like DSL, T1s, Frame Relay, ATM, etc.) we build a network using /30 (IPv4) though Cisco does support a /31 (IPv4). Often using things like RFC1918 addresses. What you are suggesting sounds more like bridging than routing and for most WAN networks is not the preferred method (bridge interfaces eat up router resources more than routed interface). Bridging doesn't scale well across the WAN.
--
Neil Cherry
Linux Home Automation
Linux HA Blog
Author: Linux Smart Homes For Dummies
|
|
RARPSL
join:1999-12-08
Suffern, NY | Thank you for your reply (and the mention in your web page of my input).
I have quickly scanned it but do not right now have time to read all of it for critiquing.
We have drifted from my original point that (due to the use of the low order 64 bits for the interface address) the number of IPv6/64 prefixes (2^64) is the equivelent of the number of possible IPv4 addresses (2^32) NOT 2^128.
I am currently not at home but once I get back I plan to pull out my O'Reilly IPv6 book to check on a number of areas.
Due to aggregation, the size of the backbone routing tables do not need to be larger than needed to get to the ISP. The ISP's tables only route over their LAN. If the ISP has 128 distribution points, each only needs to have 128 entries for LAN routing as well as Gateway entries to the backbone (I am assuming the each distribution point has it own unique prefix). The customer's prefix is only listed in the router that it is connected to so that router's table only needs to be large enough to handle the prefixes that are serviced by it as well as those of the routers it talks to. |
|
ncherryPremium
join:2003-07-13
Monroe Township, NJ | RARPSL, you are quite correct, I'm moving this quite off-topic. I'll stop here and if you ever want to discuss this privately just use my contact information here or the email address on my web page (top right). I want to thank you for the information it's given me quite a few things to think about. I've already started to make changes to my notes web page and I've got quite a few more to make. 
--
Neil Cherry
Linux Home Automation
Linux HA Blog
Author: Linux Smart Homes For Dummies
|
|
|
|
RARPSL
join:1999-12-08
Suffern, NY | said by ncherry:RARPSL, you are quite correct, I'm moving this quite off-topic. I'll stop here and if you ever want to discuss this privately just use my contact information here or the email address on my web page (top right). I want to thank you for the information it's given me quite a few things to think about. I've already started to make changes to my notes web page and I've got quite a few more to make. No problem about getting off topic. I do not see it as such. I made the statement that the correct number to compare to 2^32 was 2^64 NOT 2^128 and needed to explain/defend my comment.
I can recommend the O'Reilly book for IPv6 information.
Once I get home and have time to look at and review your page, I might contimue our discussion. |
|
ncherryPremium
join:2003-07-13
Monroe Township, NJ | said by RARPSL:No problem about getting off topic. I do not see it as such. I made the statement that the correct number to compare to 2^32 was 2^64 NOT 2^128 and needed to explain/defend my comment. I can recommend the O'Reilly book for IPv6 information. Ah, I'm kind of glad you didn't say it that way to begin with. It would have been a shorter conversation.
I have the book around here somewhere but I can't find it (just graduated with a BS, now working on an MS).
Had an interesting discussion at work with one of the engineers (he's signing up for the trial too) and we came to the conclusion that if left up to the end user they'll only use 1 subnet.
--
Neil Cherry
Linux Home Automation
Linux HA Blog
Author: Linux Smart Homes For Dummies
|
|
RARPSL
join:1999-12-08
Suffern, NY | said by ncherry:Had an interesting discussion at work with one of the engineers (he's signing up for the trial too) and we came to the conclusion that if left up to the end user they'll only use 1 subnet. I tend to agree so long as you say AVERAGE user. Most do not even have a router so the number of subnets is not an issue for them (they use the supplied IPv4/32 single address as is). Those who do have a router will not have enough devices to need a second subnet. I have seen people with both Wired and Wireless LANs who separate the two into different subnetworks for control purposes. As I said, the ISP giving out /56s (in lieu of a /48) will allow them to get 4096 times the number of prefixes to assign to their customers while still allowing the customer to have the same number of subnetworks that the current IPv4/32 + Class-C LAN NATing provides.
Given the size of the block the ISP is being given, this 4096 fold increase in prefixes should be a large enough supply of prefixes. |
|
ncherryPremium
join:2003-07-13
Monroe Township, NJ | Yes, forgive me I definitely meant to say average user. Fortunately the engineer I was talking to is an above average Lab-ie. |
|
