From: Brian Hescock (bhescock@xxxxxxxxx)
Date: Tue Nov 20 2001 - 22:08:43 GMT-3
Parry,
Assuming there is another equal cost path from R2 to 10.1.1.1, it would
do per packet if it's process-switching, per destination if it's
fast-switching and also per destination by default if you're doing cef.
However, there's a difference in per destination between fast-switching and
cef. In fast-switching it goes by destination address only. If you have
two paths but only one destination it would only use the one path all the
time. With cef it's per source and destination. So even though you still
only had one destination you could use both paths if you had more than one
source going to that same destination, the algorithm should direct the
packets from the other source to use the second equal cost path.
But with any of the switching methods, both equal cost paths will be used
approximately equally over time given that you normally have multiple
destinations across both paths. Where you run into a problem in some
situations is where you have a server and a back-up device (same
source/destination all the time and heavy traffic load at certain times),
it will always use the same path unless you do per packet and could lead to
bandwidth utilization issues. That's not usually a problem since backups
are typically done at night though (merely an example). Or when you only
have a couple destination networks but one is far more heavily accessed
than the other.
Brian
"Chua, Parry" wrote:
> Hello Brain,
>
> Back yo your example.
>
> > R1--R2---R3--10.1.1.1
>
> Ok, ping from R1 to 10.1.1.1 will be process switch at R1, but could be
> different when forward by R2 and R3.
>
> Ping from R2 to 10.1.1.1 will still do per packet destination, right ? (
> assume that
> R2 has multiple path to reach 10.1.1.1
>
> Parry Chua
> -----Original Message-----
> From: Brian Hescock [mailto:bhescock@cisco.com]
> Sent: Tuesday, November 20, 2001 9:37 PM
> To: Chua, Parry
> Cc: Hansang Bae; ccielab@groupstudy.com
> Subject: Re: OSPF Load Balance
>
> Parry,
> Ping doesn't do per packet load balancing per se, it's a matter of
> where you do the ping and how the router that is being traversed is
> configured. Example:
>
> R1--R2---R3--10.1.1.1
>
> Ping from R1 to 10.1.1.1. The ping packet is process-switched on R1
> since
> the packet is originated on the router. But on R2 the packet could be
> either process-switch, fast-switched, cef switched, parallel-express
> forwarding, etc, depending upon the platform and type of switching
> configured on the router / interface. In terms of load balancing, it
> reacts different with each switching path and is something that should
> be
> studied for you day-to-day operations, not just for the lab.
>
> Brian
>
> "Chua, Parry" wrote:
>
> > Hi,
> >
> > There is two way of load share ? balanceing ? in Cisco router
> > implementation, per packet and per destination. With route cache turn
> > on, per destination load share/balance is used. I know ping will do
> per
> > packet load balancing, ie one packet go interface1 and next interface2
> > and son on.
> >
> > Parry Chua
> >
> > -----Original Message-----
> > From: Hansang Bae [mailto:hbae@nyc.rr.com]
> > Sent: Tuesday, November 20, 2001 12:03 PM
> > To: ccielab@groupstudy.com
> > Subject: RE: OSPF Load Balance
> >
> > >>From: "Chua, Parry" <Parry.Chua@compaq.com>
> > >>I thought that when we show IP route, there could be one or more
> path
> > to
> > >>reach a particular destination based on the routing protocol type.
> For
> > >>OSPF, if we know there could be more than one path to reach that
> > >>destination,
> >
> > Correct.
> >
> > >>we can then manuiplate the COST to get a equal cost
> > >>for load balanceing, right ?
> >
> > Perhaps. Load Balancing is not as easy as people think.
> >
> > >>Other protocolos such as (E)IGRP support
> > >>unequal cost load balancing. As for RIP, I belive we can use offset
> to
> > >>change the hop count from unequal to
> > >>equal and vice versa, right ?
> >
> > But keep in mind that fast processing's job is to limit the lookup of
> > the
> > routing table (forwarding table I guess). So even if you have two
> equal
> >
> > cost routes to a destination, by default, it will use one and only one
> > link. If it didn't, it kinda defeats the purpose of fast-processing.
> > If
> > you do a "sho ip cache" you'll see what I mean.
> >
> > hsb
This archive was generated by hypermail 2.1.4 : Fri Jun 21 2002 - 06:45:19 GMT-3