RE: How to adjust the "traffic share count/ratio" of equal cost IGP path ?

From: Andy Li (ali@cisco.com)
Date: Tue Sep 26 2000 - 09:07:01 EDT


Chris

  Have you drived CSCO lately?

  Juniper's fast re-route can drop 20,000+ packets while Cisco's FRR drops
less than 20. JNPR needs one LSP to backup the other while CSCO can use one
to backup many.

Best
Andy
-----Original Message-----
From: Martin, Christian [mailto:cmartin@gnilink.net]
Sent: Tuesday, September 26, 2000 4:38 AM
To: 'Yu Ning'; Andy Li; cisco-nsp@puck.nether.net
Cc: Chinanet-vt
Subject: RE: How to adjust the "traffic share count/ratio" of equal cost IGP
path ?

Yu,

The router, in general, is attempting to find one and only one path to a
destination - the best path. If all things are equal in regards to the
particular protocols path selection algorithm, then a Cisco will
load-balance among up to 6 equal cost paths. If your links are of different
bandwidth/delay/cost - if their metrics differ, then they should not be part
of this equal cost balancing, because they violate the very premise of
equality itself. This is why IGPs do not allow for very robust traffic
engineering. BGP can do a little more, but the configuration nightmare is
to be avoided. This brings us to MPLS.

In a nutshell, MPLS allows one to choose an arbitrary path through the
network that uses an arbitrary (acyclic) sequence of node,link pairs to form
a Label Switched Path. You can choose the path based on an administrative
'color', the amount of bandwidth needed, the level of QoS based on IP ToS,
etc. Since the LSPs are set up before traffic flows, and because Label
Switching Routers merely have to look up label mappings in a table, the
forwarding is fast - even when only partially hardware accelerated - at all
available interface speeds. In the Cisco world, there is still some legacy
nomenclature (Tag Switching), but there is full support for MPLS, and it
interpoerates well with Juniper (although jnpr does MPLS better - no fast
reroute in Cisco yet.)

If you don't want to use EIGRP, then look at MPLS. It may help you out a
bunch.

./chris

-----Original Message-----
From: Yu Ning [mailto:yuning@ns.chinanet.cn.net]
Sent: Tuesday, September 26, 2000 1:38 AM
To: Andy Li; cisco-nsp@puck.nether.net
Cc: Chinanet-vt
Subject: Re: How to adjust the "traffic share count/ratio" of equal cost IGP
path ?

Hi Andy,

Here is the result:

rtr4-c-1-bjbj#sh ip cef 202.97.9.60
202.97.9.60/30, version 35751988, per-destination sharing, 0 packets, 0
bytes
  via 202.97.9.17, POS2/0, 21 dependencies
    traffic share 1
    next hop 202.97.9.17, POS2/0
    valid adjacency
  via 202.97.9.161, POS2/3, 21 dependencies
    traffic share 1, current path
    next hop 202.97.9.161, POS2/3
    valid adjacency
  via 202.97.10.173, ATM4/1.1, 22 dependencies
    traffic share 1
    next hop 202.97.10.173, ATM4/1.1
    valid adjacency
  0 packets, 0 bytes switched through the prefix

rtr4-c-1-bjbj#sh ip cef 202.97.9.60 int
202.97.9.60/30, version 35751988, per-destination sharing, 0 packets, 0
bytes
  via 202.97.9.17, POS2/0, 21 dependencies
    traffic share 1
    next hop 202.97.9.17, POS2/0
    valid adjacency
  via 202.97.9.161, POS2/3, 21 dependencies
    traffic share 1, current path
    next hop 202.97.9.161, POS2/3
    valid adjacency
  via 202.97.10.173, ATM4/1.1, 22 dependencies
    traffic share 1
    next hop 202.97.10.173, ATM4/1.1
    valid adjacency

  0 packets, 0 bytes switched through the prefix
  Load distribution: 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 (refcount 65)

  Hash OK Interface Address Packets
  1 Y POS2/0 point2point 0
  2 Y POS2/3 point2point 0
  3 Y ATM4/1.1 point2point 0
  4 Y POS2/0 point2point 0
  5 Y POS2/3 point2point 0
  6 Y ATM4/1.1 point2point 0
  7 Y POS2/0 point2point 0
  8 Y POS2/3 point2point 0
  9 Y ATM4/1.1 point2point 0
  10 Y POS2/0 point2point 0
  11 Y POS2/3 point2point 0
  12 Y ATM4/1.1 point2point 0
  13 Y POS2/0 point2point 0
  14 Y POS2/3 point2point 0
  15 Y ATM4/1.1 point2point 0
  16 Y POS2/0 point2point 0

Just like in the "sh ip ro", all the three links have equal share of the
traffic.
Because of them is ATM PVC, can I set him to share a small amount of traffic
?
Unequal share ? Note we'd not use MPLS tunnel tricks.

regards,

------------------------------------------
(Mr.) Yu(2) Ning(2)
Support Engineer, Int'l/Domestic Routing
ChinaNET (AS4134) Backbone Operation
Beijing, P.R.C. +86-10-66418105/8121/8122
------------------------------------------
----- Original Message -----
From: "Andy Li" <ali@cisco.com>
To: "Yu Ning" <yuning@ns.chinanet.cn.net>; <cisco-nsp@puck.nether.net>
Cc: "Chinanet-vt" <Chinanet-vt@cisco.com>
Sent: Monday, September 25, 2000 11:56 PM
Subject: RE: How to adjust the "traffic share count/ratio" of equal cost IGP
path ?

> Please give theo utput of the following command
>
> sh ip cef 202.97.9.60
>
> Thanks
> Andy
>
> -----Original Message-----
> From: Yu Ning [mailto:yuning@ns.chinanet.cn.net]
> Sent: Monday, September 25, 2000 12:21 AM
> To: cisco-nsp@puck.nether.net
> Cc: Chinanet-vt
> Subject: How to adjust the "traffic share count/ratio" of equal cost IGP
> path ?
>
>
> Hi All,
>
> We know that if we see equal cost IGP routes, the Cisco box will load
> balance between
> the different output interfaces. For example:
>
> Routing entry for 202.97.9.60/30
> Known via "isis", distance 115, metric 5, type level-2
> Redistributing via isis
> Last update from 202.97.10.173 on ATM4/1.1, 5d04h ago
> Routing Descriptor Blocks:
> * 202.97.9.17, from 202.97.9.17, via POS2/0
> Route metric is 5, traffic share count is 1
> 202.97.9.161, from 202.97.9.161, via POS2/3
> Route metric is 5, traffic share count is 1
> 202.97.10.173, from 202.97.10.173, via ATM4/1.1
> Route metric is 5, traffic share count is 1
>
> The "*" sign indicate the current route used. But normally the traffic
load
> will evenly
> shared in all the parallel paths, no matter how different BW they are.
Then
> I wonder if
> there is any method to adjust the "traffic share count" of the different
> output interface,
> so that I can get an uneven load share?
>
> The possibility of this idea comes to me when I dig some document on cisco
> MPLS-TE, it says
> that if two MPLS tunnels have different BW, the "traffic share count" will
> be adjusted according
> to the bandwidth ratio automatically. Because the load share is based on
the
> CEF, and is generic
> to all kinds of IGP, then I think it maybe possible in ISIS.
>
> Our environment is: GSR with dCEF enable, IOS 12.0 train. IGP=ISIS.
>
> Any input? Thanks!
>
>
> Yu Ning
> -------------------------------------------
> (Mr.) Yu(2) Ning(2)
> ChinaNet Backbone Operation
> Networking Dep.,Datacom Bureau
> China Telecom.,Beijing,P.R.C
> +86-10-66418105/66418121/66418122
> -------------------------------------------
>
>



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:17 EDT