2011.05.27 петък 19:42:09 Russ Allbery: > "Тони Стоев | Toni Stoev" <[email protected]> writes: > > > Why so, what is the gain? A promising example: use of arbitrary-length > > numbers in network addresses. Node addresses can be adequate in size to > > fit local needs, not squeezing the universal expansion of the network. > > Just to add a bit to what Ian said, please note that methods of writing > down arbitrary-length numbers in binary are well-understood and have been > well-understood for quite some time. Lack of an algorithm is not why > network addresses use fixed-width fields. Rather, the problem is that > arbitrary-length numbers are much slower because you don't know exactly > how long the packet header is going to be and therefore cannot write > highly-optimized software (generally accompanied by highly-optimized > hardware) for routing packets when you use variable-length packet headers. > The performance penalty would be prohibitive for Internet routers.
Unless numbers and the header are processed at the very moment they come in, properly like with hashing. This is greatly a matter of hardware. But as a beginning the software can lead the way. Furthermore, network packet processing depends on protocol layering. Here is the solution to that: Role-Based Architecture: http://isi.edu/newarch/DOCUMENTS/hotrba.paper.pdf Processing packets based on functions dynamically identified in header. > This was a topic of considerable discussion during the IPv6 design, since > of course everyone would have preferred to use arbitrary-length addresses > had it been possible. Not enough decision makers could have imagined possible the depth and generality of the solutions we are discussing with you. > -- > Russ Allbery ([email protected]) <http://www.eyrie.org/~eagle/> > > > -- > To UNSUBSCRIBE, email to [email protected] > with a subject of "unsubscribe". Trouble? Contact [email protected] > Archive: http://lists.debian.org/[email protected] > > > -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected] Archive: http://lists.debian.org/[email protected]
