>Sorry Howard, I appear to have deleted the original
>thread.
>
>Howard,
> You mentioned that computing the network mask from
>a 6 bit field would be detremental causing unnecessary
>CPU overhead, however, this CPU overhead would only be
>in nano-seconds compared to the serialisation delay
>i.e getting the extra bits on and off the wire.
The issue isn't serialization delay. Masking has to take place in
the "fast path" of the router, which indeed may be an ASIC, etc.,
that only knows how to do very basic operations.
Every nanosecond you add to a packet in the main data path is
significant, especially at gigabit speeds. The basic operations in
this path (I'm talking about pure forwarding--don't even think about
routing protocols) include:
extracting the destination field, masking it, and looking up
the next hop
decrementing the TTL field and recomputing the header checksum
adding whatever internal headers are needed and sending it to the
next hop.
While fragmentation is less and less a requirement, there are more
and more needs to do traffic shaping and other QoS stuff. Lots of
carriers want accounting in the fast path.
Protection against denial of service means either filtering (ouch!)
or reverse path verification, which is somewhat more scalable but
still takes cycles.
>
>As a counter argument the first byte of an IP header
>is generally 0x45, whereby the first 4 bits represent
>the IP version and the least significant 4 bits
>represent the number of 32 bit quantities making up
>the IP header.
And you're talking about a header that was designed no later than
1981, and even then had some backwards compatibility issues. While
it is true that the first few bits of the IPv6 header are a version
indicator, that can be tested with a single instruction.
IPv6 doesn't have a header length field. Instead, it has a fixed
basic header with a pointer to optional fixed-length extended
headers, or a null pointer that says there are no more header fields.
This was no accident; it was very carefully considered as necessary
for high-performance routing.
>
>Just a thought,
>
>Phil.
_________________________________
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]