[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: DSTM





Iljitsch van Beijnum wrote:

On 24-jun-04, at 18:47, Fred Templin wrote:

Iljitsch van Beijnum wrote:


IPv4 over it where needed as soon as the necessary mechanisms (IPv4-in-IPv6 tunneling, IPv6 DNS resolver discovery) are available and all the hardware can handle IPv6 in the fast path. (I had a few conversations with vendors at the summit thing last week and many still do IPv6 in software.)


IPv6 flows are identified by the 3-tuple of (src, dst, flow_id), i.e.,
there are 128+128+20=276 bits needed for handling IPv6 in the
fast path. Is anyone making CAMs that wide?


Wouldn't just src + flowid be enough? And not everyone supports initializing the flow label to something else than 0... So it's back to port numbers in that case.


Dunno, but I don't think it can be assumed that port numbers
will be accessible (other than at the endpoints, that is).

Fred
ftemplin@iprg.nokia.com


But you are assuming the fast path would be some kind of flow switching. I wish you wouldn't, as this doesn't work in ISP environments where the number of new flows per second is just too high, not to mention what happens when worms spit out packets with random destinations at high speed.