[c-nsp] Troubleshooting Lag between GigE interfaces

[Rodney Dunn]

> > My advice to customers is you do it one of two ways:
> > 
> > a) Get an RSP fast enough to run in centralized CEF mode
> > or
> > b) Make sure all your cards are dCEF capable and have
> >    the CPU power to handle the features and run in dCEF mode.
> 
> Or:
> 
> c) Use a reasonable fast RSP (4 upwards) and use VIPs for all
>    higher-speed interfaces, and maybe reuse some old SSIP/FSIP
>    for slow-speed interfaces you don't need those dCEF-only
>    features on.
> 
> I see no reason for the OP to throw out the FSIP as you suggested,
> because his problem was disabled dCEF on the VIP^WGEIP due to lack
> of RAM on it. An SSIP/FSIP can't make an RSP4 sweat when using CEF.

That's correct.  But the problem is you have to understand
the underlying switching vectors and how they work.  If
you said you have all packets going from a dCEF capable card
to a dCEF cabpable card and never between a dCEF capable card
and a non dCEF capable card I agree with you.

Honestly, all permutations should work.  I'm just telling you
the reality is that when you use a bunch of different cards
for this particular platform you make the switching vectors
much more complex and it opens the door to more bugs.

Notice I didn't say it shouldn't work. :)
I just wouldn't do it.


> 
> > If I have an RSP that can do features at X rate which
> > is more than the cards can do at Y rate I'm better off
> > in centralized RSP mode.  If the VIPs can do the work
> > distributed much faster than I can do it centrally then
> > I should go with dCEF mode.
> > 
> > I see the most problems when customers try to run it
> > in some hybrid mode.
> 
> As long as they know the constraints, this is no problem at all.
> The problem is missing knowhow, not a technical problem. It's a
> problem of not "using the right tool in the right way for the
> job" -- I guess you'll agree here. :-)

Actually there are technical problems that are really bugs.
I see it too many times where a customer wants to know why
feature X that is only supported with dCEF doesn't work with
a legacy IP board that doesn't support it.
Not to mention that with the legacy IP boards you have zero ability
to troubleshoot what's actually happening on the board itself.

Don't get me wrong in my response here.  There are a lot of situations
where having legacy ip boards and non legacy ip boards works just
fine.  I just don't recommend it.  Even if he had enough memory
to do dCEF on the GIG cards he couldn't dCEF switch traffic to the
legacy IP boards so most likely we would have had to troubleshoot
that too. :)

> 
> BTW, thanks for your very valuable contributions to this mailing
> list here, appreciated!

Glad to help.

> 
> 
> Best regards,
> Daniel
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/