[nsp] ip load-sharing per-packet - cef accelerated ?

[Oliver Boehmer (oboehmer)]

All,

just to clear this up (some wrong information has been floating around in this thread):

"ip load-sharing per-packet" is CEF accelerated, there is no significant overhead when it comes to router performance compared to cef per-destination (I'm now only talking about non-GSR systems). The sole difference between the two is how we select one of the 16 loadsharing bucket when we switch a packet.

cef per-destination (the default loadsharing method) uses a hash with <source,dest>, so the name is a bit misleading. It does NOT balance per destination address only, as someone mentioned. And it also does not load-balance per micro-flow as L4 information or protocol# is not used while building the hash. 

There is a command "show ip cef exact-route <source> <dest>" which can be used to find which path a given session (<source,dest>) will take:

r1#sh ip cef 2.0.0.0 internal
2.0.0.0/8, version 14, attached, per-destination sharing
0 packets, 0 bytes
  via Serial2/0, 0 dependencies
    traffic share 1
    valid adjacency
  via Serial3/0, 0 dependencies
    traffic share 1
    valid adjacency

  0 packets, 0 bytes switched through the prefix
  tmstats: external 0 packets, 0 bytes
           internal 0 packets, 0 bytes
  Load distribution: 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 (refcount 1)

  Hash  OK  Interface                 Address         Packets
  1     Y   Serial2/0                 point2point           0
  2     Y   Serial3/0                 point2point           0
  3     Y   Serial2/0                 point2point           0
  4     Y   Serial3/0                 point2point           0
  5     Y   Serial2/0                 point2point           0
  6     Y   Serial3/0                 point2point           0
  7     Y   Serial2/0                 point2point           0
  8     Y   Serial3/0                 point2point           0
  9     Y   Serial2/0                 point2point           0
  10    Y   Serial3/0                 point2point           0
  11    Y   Serial2/0                 point2point           0
  12    Y   Serial3/0                 point2point           0
  13    Y   Serial2/0                 point2point           0
  14    Y   Serial3/0                 point2point           0
  15    Y   Serial2/0                 point2point           0
  16    Y   Serial3/0                 point2point           0
r1#
r1#sh ip cef exact-route 1.1.1.1 2.2.2.2
1.1.1.1         -> 2.2.2.2        : Serial3/0 (attached)
r1#sh ip cef exact-route 1.1.1.2 2.2.2.2
1.1.1.2         -> 2.2.2.2        : Serial2/0 (attached)
r1#sh ip cef exact-route 1.1.1.3 2.2.2.2
1.1.1.3         -> 2.2.2.2        : Serial2/0 (attached)
r1#

Please also check http://www.cisco.com/univercd/cc/td/doc/product/software/ios122/122cgcr/fswtch_c/swprt1/xcfcefc.htm#1000944

	oli


> -----Original Message-----
> From: Dan Hollis [mailto:goemon at anime.net] 
> Sent: Dienstag, 11. März 2003 01:07
> To: Rob Thomas
> Cc: Cisco List
> Subject: Re: [nsp] ip load-sharing per-packet - cef accelerated ?
> 
> 
> On Mon, 10 Mar 2003, Rob Thomas wrote:
> > ] My understanding is that 'ip load-sharing per-packet' is 
> CEF accelerated,
> > ] yes?
> > No.  It is per packet load balancing instead of per flow 
> load balancing.
> > This puts quite a bit of additional load on the router, and 
> may result
> > in out of order packets arriving at the destination.  I recommend
> > against it.
> 
> Are you sure per-packet isn't CEF accelerated?
> 
http://www.cisco.com/univercd/cc/td/doc/product/software/ios121/121cgcr/switch_c/xcprt2/xcdcefc.htm

Why would it put additional load on the router? Every packet does a CEF 
lookup anyway, yes?

Has anyone on this list actually done an analysis of CPU usage for CEF 
per-packet vs CEF per-destination?

-Dan
-- 
[-] Omae no subete no kichi wa ore no mono da. [-]

_______________________________________________
cisco-nsp mailing list  cisco-nsp at puck.nether.net
http://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/