In einer eMail vom 28.11.2007 23:59:17 Westeuropäische Normalzeit schreibt [EMAIL PROTECTED]:
On Nov 28, 2007, at 2:51 PM, [EMAIL PROTECTED] (mailto:[EMAIL PROTECTED]) wrote: A single matrix lookup for any destination which is still in a different geo-patch can also be done by ASIC. True, that should be a straightforward array lookup. Can anybody tell the difference in time compared with prefix lookup by ASICs ? Yes, the longest-match would be much, much slower. ;-) In any non-trivial forwarding table, the longest-match will require multiple memory reads. Exactly how many is very much a function of the specific lookup algorithm. In a naive v4 implementation, you could reasonably take around 24 memory reads. However, the array lookup should be just a single read. Tony = You see, this concept is not only scalable but enables faster forwarding as well Another point: Mobile IP-phones. They can provide the geo-location down to the second (rather just down to the degree). If they move to a different 1x1 geo-patch they have to notify their base stations about the new long/lat degrees and advertise the changed reachability only within the affected 1x1 square degree geo-patches, within the previous (for withdraw) and within the new one. No world-wide routing churn. I am sure, if each of you started to trust your own eyes, you would discover many more improvements than me. You know the hurdles and the ackwards of the current BGP better than me. Heiner Heiner
