1 2 3 4 Previous Next 53 Replies Latest reply: Apr 13, 2012 9:00 AM by Greg, CCNP, JNCIP Go to original post RSS
      • 30. Re: My company is getting a /34 IPv6 prefix
        Greg, CCNP, JNCIP

        The problem with everyone having their own address space, which are not subnets of their ISP's big address blocks, is that the global routing table grows as more people connect their networks to the Internet.  It's a big problem for ISPs, because they have to keep upgrading their routers to have enough memory to hold the routing table.


        ISPs would prefer if their customers would get addresses from their ISPs, because that minimizes routing table growth.  For people who are multihomed, that means getting separate address block from each ISP, and configuring an address from each of those blocks on every host/interface.


        Maybe someday we'll figure out a way to grow the Internet more efficiently, AND save people from ever having to renumber.

        • 31. Re: My company is getting a /34 IPv6 prefix
          Joshua Johnson - CCNP R&S

          IPv6s automatic renumbering mechanisms will hopefully mature quickly....


          And as for my last question, Sir?

          • 32. Re: My company is getting a /34 IPv6 prefix
            Greg, CCNP, JNCIP

            Yes, assigning multiple addresses per host/interface was supposed to ease the pain of switching ISPs, in a world where everyone gets their addresses from their ISPs.  I guess the pain is eased if there are good tools/methods for addressing hosts/interfaces.


            There might be a BGP multihoming problem w/ customers using ISP-assigned addresses, if ISPs refuse to advertise subnets of their own address blocks to other ISPs.  It depends on the agreements between ISPs - which IPv6 routes should/shouldn't be advertised?  I'm not sure about the general practice now; a couple of years ago there seemed to be movement toward advertising up to /48, regardless of if it was ISP-assigned.



            • 33. Re: My company is getting a /34 IPv6 prefix
              Joshua Johnson - CCNP R&S

              We actually obtained our /30 from ARIN... I wasn't part of the purchase but I'm assuming they checked it out with our providers.

              • 34. Re: My company is getting a /34 IPv6 prefix
                Joshua Johnson - CCNP R&S

                Thanks for the link i'll check that out.

                • 35. Re: My company is getting a /34 IPv6 prefix
                  Joshua Johnson - CCNP R&S

                  Check this out, from the same white-paper from ATT...  I know what they are talking about, but technically if the address pool is 2001:DB:1234::/48, then the range would be 2001:DB:1234::/48 - 2001:DB:1234:FFFF:FFFF:FFFF:FFFF:FFFF/48.


                  If using the industry best practice then the fourth quartet would be used for subnetting of course...


                  For the public facing networks, companies must select a group of 64-bit prefixes from their overall IPv6 address pool to assign to the public network segments such as the DMZ, CE-Firewall LAN, etc. This group of prefixes should be chosen from either the upper or the lower bounds of the IPv6 address pool. For instance if the assigned address pool is 2001:DB:1234::/48, then choose the prefixes from the 2001:DB:1234:: or the 2001:DB:FFFF::.

                  • 36. Re: My company is getting a /34 IPv6 prefix
                    Greg, CCNP, JNCIP

                    Good catch, easy to make those kinds of mistakes in IPv6 addressing, and easy to miss it.  I had to read your post a couple of times before I saw the problem.  Should've been 2001:DB:1234:FFFF::/64.  They probably also meant to use 2001:DB8 (http://tools.ietf.org/html/rfc3849).

                    • 37. Re: My company is getting a /34 IPv6 prefix

                      I work for a large ISP and put our commercial IPv6 architecture in place. We don't necessarily prefer customers to get space from us. Actually Provider Independent space (PI) is something we actively encourage.


                      Firstly if you get space from an ISP you don't need to get more space from another ISP to do multi-homing. In fact if you do it's not true multi-homing - per se. What you do need to do is have at least a IPv6 /48 allocated from ISP-A and have them give a Letter of Authority (LOA) to ISP-B to allow them to advertise it also. This is the same as with IPv4 ( it would have to be a /24 or larger) and I'll give the example here using IPv4 to make it easier.


                      This however does increase the global route table size and has added to the IPv4 address-table size problems seen historically. What happens here is that ISP-A will most likely have a larger block - say a /22 so when ISP-B advertises the /24 this will be preferred as longest match regardless of any BGP attribute outside of ISP-A's AS. In order for any traffic to route into the customer via ISP-A they must advertise the /24 also along with their /22. Once this happens the customer can do pre-pending to manipulate the routing further etc etc. So what's happened here is that the global route table just got 2 /24 routes added to it for one /24 allocation that was previously represented by the /22.


                      If customers get a PI allocation then they same thing needs to happen so PI really doesn't cause any greater increase in the global route table for multi-homing.


                      Due to this we take a customer-first policy and actively promote PI for all BGP customers but especially for IPv6. Why? Well because it's in the best interest of the customer to have PI space. We don't want our single homed customers to have to re-ip if they want to do multi-homing with another provider that won't co-operate (most will though). Also it may sound counter productive but we don't want to force customers to stay with us because it would be too much hassle or cost too much to re-ip their LAN etc. We want them to stay because they like us.:) Plus unhappy customers tell other people they're unhappy.


                      In our opinion we want to encourage the vendors to find inovative and cost effective means to keep the price of core hardware down and at the same time ensure memory and processing power evolve to facilitate the projected IPv6 table size.

                      • 38. Re: My company is getting a /34 IPv6 prefix
                        Greg, CCNP, JNCIP

                        Righto, I should've said "some ISPs prefer their customers to get addresses from ISPs..."   Also, I doubt that ISP-assigned addresses have ever been an evil plot to keep customers locked into an ISP - I think it was always a routing table growth and economics issue.  Global routing table growth is steadily positive, but business growth can be flat/negative.  It sucks to be forced to spend $Millions to upgrade one's network to stay ahead of the routing table, when business doesn't otherwise justify the expense (layoff time!).

                        • 39. Re: My company is getting a /34 IPv6 prefix
                          C1SC0M - CCNP,CWNA,Net+

                          I haven't read all the thread so I will assume ok?   If you have 2 ISPs regardless if you are using IPv4 or v6 addresses, each ISP will assign you a block of public(global) addresses.  They will only advertise to the rest of the world their own address space.  ISP-A will advertise ISP-A global addresses only and ISP-B will advertise ISP-B addresses. I don't want to enter much detail here, but the thing is that if you have your own address space and an AS identity assigned by IANA for example, you can advertise that AS by both ISPs, thus any services(WWW, EMAIL,etc.) relying on that address space will continue to provide access in case of one ISP failure.  Whenever your company decide to change any of the ISP you continue to have your own address space and it is a matter of advertise it through the new ISP. 

                          • 40. Re: My company is getting a /34 IPv6 prefix

                            Sorry C1 but we do advertise space assigned to customers by other ISP's - with their consent (LOA) - and this is common. When you are transitioning from one ISP to two you just need to ask the new ISP for the /30 to do the eBGP peering over and get your existing ISP to send LOA to the second ISP. Often the ISP will also require the address space to be in something like RADDB.  They don't need to give you any new space - they can and you can route both to both ISP's but you don't need to.


                            After you do this you can advertise this /24 or larger to the new ISP and they will/should advertise it upstream. They will however only be able to advertise this specific route and not the larger /xx block that the other ISP has assigned to them from ARIN etc - so the existing ISP must also advertise the specific block. If not all incoming traffic will come via the new ISP peering regardless of AS pre-pending. Once the routes are equal length you can pre-pend but you will also need to get the ISP's to adjust Local-preference on their AS if you want to change routes sourced directly from them or other customers of them. Most good ISP's will offer you community strings so you can do this yourself automagically.


                            We have many customers that do this and we do recommend that they get PI space but there are many roadblocks to this, mainly that it costs money and secondly that they would have to re-ip their internal network, nat translations, DNS etc etc.

                            • 41. Re: My company is getting a /34 IPv6 prefix

                              I agree Greg,


                              I don't think there is an evil plot but if a customer is unhappy and wants to move ISP but can't because it's too costly to do so then they will feel like and ARE locked in and will only have negative things to say about the ISP's service. It does **** to have to upgrade here but business is business and you have to spend money to make money . It's possible to architect a network so you don't need to have full routes everywhere but it won't give you an optimal public routing topology and thus the service won't be as good as those ISP's that do.


                              Do you know where the full-internet route table starts in a traceroute over your ISP? It's not so easy to tell for the average customer.

                              • 42. Re: My company is getting a /34 IPv6 prefix

                                Obviously this is all dependant on having a /24 or /48 allocation in the first place but multi-homing is good enough justification for most. If you have a smaller block you won't be able to do this and you'll have to use different subnets per ISP and do some creative dual-stack ip addressing or get some PI space and re-ip or NAT using that.

                                • 43. Re: My company is getting a /34 IPv6 prefix
                                  Greg, CCNP, JNCIP

                                  "Do you know where the full-internet route table starts in a traceroute over your ISP?"


                                  I don't understand the relevance of this question, and the premise sounds questionable.  Are you implying that it's possible to tell, from traceroute alone, which of the hops shown has a full Internet routing table?


                                  Traceroute doesn't tell you that.  All you can determine from traceroute is: 1) that each hop which sends you an ICMP time-exceeded actually decrements IP TTL and has a routing table entry back to your IP address, 2) that the previous hop's routing table indicated that the current hop is its best path to the destination, and 3) that no firewalls prevented your probes out nor each hop's replies back to you (if you got a reply).

                                  • 44. Re: My company is getting a /34 IPv6 prefix

                                    The relevance is in relation to optimal routing path. i.e. does it take 10 hops to get from A to B or 5.


                                    As far as where the full route table starts in the path -- well yes you can't tell this from a traceroute alone - which was exactly my point sir - I guest I was pointing out that most customers just know how to do traceroute and ping etc and wouldn't be able to do the additional technical investegation required that could ( depending on the topology and available information) show that there wasn't a full route table present for a few router hops into the ISP's network.


                                    I've seen some large customers do this though and base their ISP choice accordingly. There's usually an underlying reason for this such as - customer is running their own VoIP service or have a latency sensitive app running over this.