[c-nsp] disabling dCEF per VIP on 7500?

Rodney Dunn rodunn at cisco.com
Thu Aug 12 19:39:29 EDT 2004


Gert,

No way to do it that I know of.
If I remember correctly it's because of the way
the FIB is downloaded to the VIP's.  It's not
unicast to each VIP from what I recall.

And honestly, I would never run a 75xx that way.
It makes the switching paths too complicated and
open to bugs.

If you go to any of the fancy QOS features we
only support them in full dCEF mode because of
multiple reasons: scalability, simplicity in the
swithing vector, etc.

Rodney

On Thu, Aug 12, 2004 at 09:37:45PM +0200, Gert Doering wrote:
> Hi,
> 
> I've searched for a while in the documentation, but can't find anything
> that does what I want - well, maybe it's not possible, but it's worth a
> try asking here.
> 
> I have a 7507 with 12.0(27)S2 on it, and 5 VIPs in it.  Two are VIP2-50s
> with 128Mb DRAM and "fastish" interfaces (POS-OC3, FE), the other 
> three are VIP2-40's with 64 Mb DRAM and "slowish" interfaces (10Mbit 
> Ethernet, 2Mbit G.703 or X.21).  
> 
> RSP4 with 256 Mb DRAM, full BGP tables.
> 
> Until recently, the box was running "plain CEF" (central CPU) and did
> well.  Some CPU spikes here and there, but nothing critical, and no urgent
> need to change anything.
> 
> Nevertheless I decided to go to distributed CEF, because it's supposed
> to scale better in case of "many packets flowing" in parallel to "large
> BGP updates".
> 
> Now the interesting part is: I want to have the box run dCEF on the 
> VIP2-50s, but not on the VIP2-40s.
> 
> I know I can disable dCEF per-interface ("no ip route-cache distributed"),
> but it's not disabling the actual FIB download onto the VIPs.
> 
> The VIP2-40s are very tight on memory right now:
> 
> VIP-Slot1>sh mem
>                Head     Total(b)     Used(b)     Free(b)   Lowest(b) Largest(b)
> Processor   60B57F20    55214304    53458740     1755564     1721604  1645500
>       PCI   30000000     2097172     2041668       55504       55504    28468
> 
> ... and I'm fairly sure that in a couple of months, their memory will
> be full (this is with all interfaces on "no ip route-cache dist" already,
> didn't make a difference in memory usage).
> 
> Documentation says something like "in that case, dCEF will just be 
> disabled on the offending line card", but from experience, I assume that
> it will result in funny things (CBUS complex, VIP2 restarts, or some
> other nasty service interrupting surprise).  So I want to turn off 
> dCEF for that whole line card *now*.
> 
> --> Is that possible?  Disable dCEF on one VIP purposely, and have it
> running on the other ones?
> 
> gert
> 
> -- 
> USENET is *not* the non-clickable part of WWW!
>                                                            //www.muc.de/~gert/
> Gert Doering - Munich, Germany                             gert at greenie.muc.de
> fax: +49-89-35655025                        gert at net.informatik.tu-muenchen.de
> _______________________________________________
> 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/


More information about the cisco-nsp mailing list