Curtis Villamizar <curtis(a)ans.net> writes:
> There are routers that handle {LS}SRR at near full speed. Stuffing it
> on the broken path in is a bug. Fast-path/broken-path
> is just a bug.
I'm not sure I agree, particularly in the case where
"broken-path" involves delegating work to another
processor.
I will agree with you that the fast switching scheme in
current IOS is gross, however it was expedient, and is the
reason there are so many DS3 networks operational today.
Remember that prior to 10.2-viktor it was much worse...
The fact that one fast-path/slow-path scheme is broken in
several subtle ways does not really convince me that it's
a bad idea in general. In particular, when we're dealing
with the sorts of time budgets needed to support multiple
OC48s per card, while _I_ would love to have a single path
and might be willing to live with reduced port-density
(clever secondary-circuit wiring, for example) to avoid
running too close to per-packet time-budgets, other people
might convincingly argue that if one can determine in
advance that drops will happen if a particular datagram is
fully processed, and the combination of determining that
and delegating the work to another processor falls within
the time budget under bad-case to worst-case conditions,
it's a good idea to have multi-path forwarding.
So, yes, simplicity is good and easy to diagnose.
Unfortunately, when the diagnosis is a simple "you haven't
got enough processing power to handle your traffic pattern
using a single path", forwarding-path simplicity is one of
the things that needs evaluation.
Of course, I am also from the school of thought that
someone blasting a huge amount of worst-case traffic on
one interface should never affect normal traffic across
that and especially not other interfaces. That school
doesn't like the thought of not being able to handle a
full OC-12's worth of offered traffic on an interface
because someone is jamming a large fraction of that
bandwidth worth of Christmas-tree packets at someone's
router.
Discovering that routers X handle Y at "near full speed"
always worries me. Consequently, I worry alot. :-)
Sean.
