Let me try to inject some perspective. This thread seems to have an underlying
premise about certain platform design goals being universal across all router
applications.
Fortunately or unfortunately, the requirements for a SOHO router, a branch office
router, an enterprise core router, an ISP POP router, and an ISP core router have
substantially different motivations.
Take the small routers at SOHO or branch. These devices are price sensitive. If both
lookups and forwarding can be done in the same "cheap" processor, how more affordable
is the router if it doesn't need a separate RP?
There tends to be a lot of assumptions that routers present significant internal delay
(and a salesish assumption that "layer 3 switches" somehow don't). In most
circumstances, you'll find router-associated delays, below the major ISP level, to be
more a matter of serializatin and queueing delay than internal transfer.
>
>
>Hey Group,
> Me again. I'm reading for my CIT and am at the section where it
>goes
>into detail of the various switching methods in the router (i.e.,
>silicon,
>CEF, autonomous, etc.) I understand how all this works and understand
>how the
>SP takes a lot of the stress away from the RP and this is good because
>your
>avoiding bogging the RP/CPU down. I have a problem with these statements
>though and want some clarification...
Cost may be more important than bogging.
>
>Taken form the book (Lammle's CIT p. 173):
>
> "This is just another reason why switching is such a good practice.
>Why
>burden the RP with every packet if it's not necessary? By using
>switching
>methods, the RP is free to use valuable CPU time on more important
>things
>than doing route lookups for every packet that comes in the router."
First, route lookup isn't terribly CPU expensive, especially if it uses an algorithm
such as the CEF mtrie. You're talking on the order of 4-5 memory accesses in CEF.
More in other methods, but not a major source.
Rewriting packet headers (e.g., for fragmentation) is much more CPU expensive.
Filtering can be.
>
>
>Basically, could somebody give me a list of some other things the RP/CPU
>has
>to do other than route lookups...(I know there are access-lists and
>other CPU
>things here, I just would like a solid list to remember). Thanks team,
Memory management
SNMP
Console & telnet
Interrupt scanning
Dynamic routing
Filtering
Traffic shaping
Debug
Packet rewriting
Accounting
Encryption, tunneling, compression
...
_________________________________
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
