RE: [address-policy-wg] Policies interact
Sidetracking #2:
Savings in IPv4 space, do not carry over into IPv6. That is why we lose these savings. But the size of the routing table increases by 4 times therefore requiring 4 times as much RAM and 4 times as much time to send/receive full routes.
If this is the case we are looking at a poor implementation (memory-consumption-wise), imho. For a routing decision you don't need 32 bits for an IPv4 prefix, and you do not need 128 bits for an IPv6 prefix. My wild guess would be that the ratio is rather on the order of 1:1.5 than 1:4. [ Anyone having statistics about the average length of an IPv4 prefix? Probably in the range of (20..)21..22(..23) ] Wilfried.
Hi Wilfried,
[ Anyone having statistics about the average length of an IPv4 prefix? Probably in the range of (20..)21..22(..23) ]
you can find this information in Philip Smiths "Weekly Routing Table Report": # routing-wg-admin@ripe.net wrote: # # This is an automated weekly mailing describing the state of the Internet # Routing Table as seen from APNIC's router in Japan. # Daily listings are sent to bgp-stats@lists.apnic.net # # If you have any comments please contact Philip Smith <pfs@cisco.com>. # # Routing Table Report 04:00 +10GMT Sat 19 Mar, 2005 # #<...> # # Number of prefixes announced per prefix length (Global) # ------------------------------------------------------- # # /1:0 /2:0 /3:0 /4:0 /5:0 /6:0 # /7:0 /8:19 /9:3 /10:8 /11:14 /12:62 # /13:152 /14:305 /15:573 /16:8321 /17:2477 /18:4186 # /19:10131 /20:10825 /21:8727 /22:12051 /23:13167 /24:86110 # /25:264 /26:257 /27:110 /28:53 /29:26 /30:90 # /31:0 /32:44 Martin VM404-RIPE
On Thu, 07 Apr 2005 11:58:15 +0200, Wilfried Woeber, UniVie/ACOnet wrote:
For a routing decision you don't need 32 bits for an IPv4 prefix, and you do not need 128 bits for an IPv6 prefix. Exact. A international routing decision can be limited to the first 64 Bits. The remaining 64 Bits are some sort of ARP-replacement.
My wild guess would be that the ratio is rather on the order of 1:1.5 than 1:4. It can be 1:1 (TCAM or sparse tries) for good implementations up to 1:4 or even worse for poor implementations.
*Of course* the IPv4 approach with 32 bits has clear advantages regarding real world implementations and a worlwide unique 32 bit address is a reasonable choice. Thus IPv6 makes only sense if: - the remaining address part can be used and internationally routed - there is a clear advantage for using larger addresses - none of these is restricted as it currently is (200 customers rule, no PI) - all important IPv4 features are available with IPv6 Face it: Either IPv6 can be used with its full advantages, which means - full and true routing support for the first 64 bits - Multihoming, PI, reasonable global table processing which imposes: - using state of the art routing hardware and yes, this means spending money, like it or not, - removing unnecessary address assignment restrictions to gain the best results from the large address space, yes, this means trusting the engineers that they will be able to handle larger tables, or IPv6 will die. The market will not accept a IPv6 with address bits overhead not usable for reasonable international connectivity. The market then will stay with IPv4, which is a reasonable choice. Greetings Oliver Oliver Bartels F+E + Bartels System GmbH + 85435 Erding, Germany oliver@bartels.de + http://www.bartels.de + Tel. +49-8122-9729-0
For a routing decision you don't need 32 bits for an IPv4 prefix, and you do not need 128 bits for an IPv6 prefix. Exact. A international routing decision can be limited to the first 64 Bits. The remaining 64 Bits are some sort of ARP-replacement.
nope. folk are using /126s internally, and have igp or ibgp carrying those prefixes. of course, they also have the classic loopbacks for bgp, which can be /128s. real hardware vendors know this and don't make the same mistakes as were made in the old a/b/c days. randy
On Thu, 7 Apr 2005 06:10:55 -1000, Randy Bush wrote:
nope. folk are using /126s internally, and have igp or ibgp carrying those prefixes. of course, they also have the classic loopbacks for bgp, which can be /128s. real hardware vendors know this and don't make the same mistakes as were made in the old a/b/c days. TCAM IC's typically process 72 address lines per cycle, which means 64 bit for the address plus 8 bit policy etc.
Even if the hardware permits it, I don't think it is a good idea to put /65 and more specific prefixes into the IGP. This has nothing to do with a/b/c mistakes, you would just throw away a possible optimization path for no additional gain at all. There is a good chance some day (in the next 50 years) these additional 8 bytes will have a better use than the current "we already know where you are" adresses. Best Regards Oliver Oliver Bartels F+E + Bartels System GmbH + 85435 Erding, Germany oliver@bartels.de + http://www.bartels.de + Tel. +49-8122-9729-0
participants (4)
-
Martin Vyskocil -
Oliver Bartels -
Randy Bush -
Wilfried Woeber, UniVie/ACOnet