On Wed, 5 Mar 2003, Pekka Savola wrote:
> On Tue, 4 Mar 2003, Quality Quorum wrote: > > The problem here is software implementation of longest prefix match. > > Up to this point it was limited to a few TLAs with 48 bits which was > > quite doable in software, this draft expands it to 61 > > Note that "this" draft does not do it, it has been already done in > addr-arch-v3, which has been approved by IESG/IAB for publication to > Proposed Standard. Then we had a bunch of RFCs specifying various allocation schemes which effectively paired it down to 48 bit (in some instances 39 bit) longest prefix match. > > > and you are > > proposing 125, which is well beyond capabilities of software based > > lookups. > > "well beyond capabilities of software based lookups" seems to be clearly > false: I've run IPv6 test on Linux / BSD-based systems, which do have > exactly that and have obtained gigabit-grade results, the same as > with IPv4. > > Perhaps that applies in a very specific scenario of sw lookups only. The real life requirement for a low cost router is OC-48 with 40-byte frames, which is about half order of magnitude above hardware limitations of PCI bus. > > Btw, it's 64 and 128, respectively: you seem to be doing exactly what's > forbidden, glueing 2000::/3 in the implementation -- or do you do > 48/64/128-bit lookup for non-2000::/3 routes? It is way easier that that, roughly speaking first you do classification and then you do longest prefix match (if you are in the right TLA or in site local), or hash table lookup (if you are in multicast), or drop. The key word is "software", so it could be easily adapted to non-tectonic changes. Again, the things you are proposing may be a way to go (personally I would love to see every enterprise router to be replaced), however, I suppose that you have to clearly understand consequences of your proposal. It seems to me that, if total router replacement is not the goal, we have to establish at least some transitional period guarantees - say address assignment policies of the next 20 years would guarantee a right of existence for software implementations. > Pekka Savola "You each name yourselves king, yet the Thanks, Aleksey -------------------------------------------------------------------- IETF IPng Working Group Mailing List IPng Home Page: http://playground.sun.com/ipng FTP archive: ftp://playground.sun.com/pub/ipng Direct all administrative requests to [EMAIL PROTECTED] --------------------------------------------------------------------
