[c-nsp] Per packet Load balancing

[Amol Sapkal]

Hi,


On Tue, 7 Sep 2004 13:53:06 -0400, Rodney Dunn <rodunn at cisco.com> wrote:
> On Tue, Sep 07, 2004 at 11:08:08PM +0530, Amol Sapkal wrote:
> > On Tue, 7 Sep 2004 12:46:33 -0400, Rodney Dunn <rodunn at cisco.com> wrote:
> > > On Tue, Sep 07, 2004 at 10:09:59PM +0530, Amol Sapkal wrote:
> > > > Seems the post went in too early.
> > > > I checked after few minutes, the incoming load balancing doesnt work
> > > > now. I think I am missing something.
> > > >
> > > > This was what was done:
> > > > ip cef enabled globally.
> > > > Following command added to the physical interfaces which have the
> > > > sub-interfaces supposed to 'static load balance' the downlink traffic.
> > > >
> > > > ip load-sharing per-packet
> > > >
> > > >
> > > > What is it that I am missing?
> > >
> > > Put that command on the "ingress" interface that feeds the downstream
> > > PVC's.
> > >
> > > ie: Your links coming in from the internet.
> >
> > Ok. I  put this on the incoming interface ( to Internet) but still it
> > didnt seem to load balance. But after I re-put it on the outgoing
> > interfaces, it is load-balancing closely( but not as accurately close
> > as process switched).
> > Any reason for this?
> 
> No.  Assuming you had it configured with process switching everywhere
> there is absolutely no difference from a load sharing perspective of
> that and cef per packet.
> 
> 
Yeah, but for some reason, it is not exact and accurate as
process-switching. But it does solve the problem. Because the value
are close within a range of 10%
(like 340K and 353K or 224K and 198K)


> >
> > The hash table shows me the 2 outgoing interfaces for each IP of the /28 pool.
> > In the 'ip cef accounting' , I dont have the 'load' option.
> 
> It's hidden for no reason...just type it.
> 

Done, it worked.

> Then do 'sh ip cef <dstprefix> internal' and see if you are using
> equal amounts over the buckets.
> 
> 

Sorry for this again, but I tried to get the meaning of the terms of
the output of the command. I couldnt egt what the term bucket means
here. (The IPs are changed for obvious reasons)

my_router#sh ip cef 2.2.2.134 int
2.2.2.128/28, version 446, per-packet sharing
0 packets, 0 bytes
  Flow: AS 0, mask 28
  tag information set
    local tag: 4
  via 2.2.2.2, 0 dependencies, recursive
    traffic share 1, current path
    next hop 2.2.2.2, Serial9/1/5.3 via 2.2.2.0/30
    valid adjacency
  via 1.1.1.1, 0 dependencies, recursive
    traffic share 1
    next hop 1.1.1.1, Serial9/1/3.100 via 1.1.1.1/32
    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  Tags imposed
  1     Y   Serial9/1/5.3             point2point        6437    none
  2     Y   Serial9/1/3.100           1.1.1.1    6438    none
  3     Y   Serial9/1/5.3             point2point        6438    none
  4     Y   Serial9/1/3.100           1.1.1.1    6438    none
  5     Y   Serial9/1/5.3             point2point        6438    none
  6     Y   Serial9/1/3.100           1.1.1.1    6438    none
  7     Y   Serial9/1/5.3             point2point        6438    none
  8     Y   Serial9/1/3.100           1.1.1.1    6438    none
  9     Y   Serial9/1/5.3             point2point        6438    none
  10    Y   Serial9/1/3.100           1.1.1.1    6435    none
  11    Y   Serial9/1/5.3             point2point        6437    none
  12    Y   Serial9/1/3.100           1.1.1.1    6437    none
  13    Y   Serial9/1/5.3             point2point        6437    none
  14    Y   Serial9/1/3.100           1.1.1.1    6436    none
  15    Y   Serial9/1/5.3             point2point        6437    none
  16    Y   Serial9/1/3.100           1.1.1.1    6437    none
my_router#


> 
> >
> > Thanks - Amol
> >
> >
> > >
> > >
> > >
> > > >
> > > >
> > > >
> > > > On Tue, 7 Sep 2004 22:04:49 +0530, Amol Sapkal <amolsapkal at gmail.com> wrote:
> > > > > Thanks!
> > > > > On Tue, 7 Sep 2004 12:20:59 -0400, Rodney Dunn <rodunn at cisco.com> wrote:
> > > > > > You shouldn't do it that way.
> > > > > >
> > > > > > You can configure CEF to do per-packet load
> > > > > > balancing.
> > > > > >
> > > > > > Turn on CEF globally and then on the inbound
> > > > > > interface feeding the downstream equal cost paths
> > > > > > do "ip load-sharing per-packet".
> > > > > >
> > > > > > Don't leave it like you have it now.  You are process
> > > > > > switching all the traffic.
> > > > >
> > > > >
> > > > > This works, I checked on the client router interface. The incoming
> > > > > traffic is load balanced.
> > > > >
> > > > > Apologies for asking it here (and not reading it myself), what does
> > > > > 'ip load-sharing per-packet' actually do?
> > > > >
> > > > > >
> > > > > > Rodney
> > > > >



-- 
Warm Regds,

Amol Sapkal

--------------------------------------------------------------------
An eye for an eye makes the whole world blind 
- Mahatma Gandhi
--------------------------------------------------------------------