On Fri, Nov 25, 2005 at 12:35:27PM +0100, Mohacsi Janos wrote: [...] > You halving approach not very useful for 9K packets, which is tend to be > common on GE and 10GE.
Just an example. Let's say you need say a minimum of 8 fragments to make sure each fragment <= path MTU (M). The obvious thing to do is to send 8 fragments where 7 have the size M, and one has some size N <= M. If N < M, you could try to make some of the fragments of size M smaller, and increase N. So let's say you have some 9100 byte packet that needs to be fragmented to 1280 byte sizes. Instead of sending something like 7 x 1280, 1 x 140 (well, I'm not really account for size of frgmentation header). You can send 4 x 1138, 4 x 1137. Again, just an example to illustrate that it might be reasonable to send smaller fragments than the MTU size. And in some cases, as with further encapsulation, this might be useful. However, there would never be a reason to send fragments of size less than half of the Path MTU (IMO), since then you could send fewer fragments. Stig -------------------------------------------------------------------- IETF IPv6 working group mailing list [email protected] Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
