MPLS Layer Traffic

Chuan Liu · ‎05-19-2009

Hi NetPro,

I am using Orion NPM to monitor the MPLS WAN links. For each MPLS interface on Cisco 3845, 2 entries come up. For example, 'G0/1.2077' is the configured interface, and 'G0/1.2077-mpls layer' is the second entry shown the NFM. There are huge statistics differences between these 2 entries. In my case, the vlan 2077 is allocated 2M bandwidth. Traffic shaping is configured. The second entry(with MPLS layer) always shows over 400% link utiliztion; while the first one shows only 30% link utilization. The shaping is never seen active.

Is this link really congested? How to understand the 'mpls layer' statistics?

Are there commands that can be used on the router to check the 'mpls layer' statistics?

Any response is appreciated.

Thanks in advance.

Giuseppe Larosa · ‎05-25-2009

Hello Larry,

try to combine

sh vlans 2077

and sh mpls forwarding int g0/1.2077

it is likely that your network management software is wrong.

However, if you are using modular QoS you may need to modify the class-maps definitions to match MPLS traffic.

if for example you want to shape both IP traffic with ip precedence 3 and MPLS exp 3 you need to define a class like:

class-map exp3_ipprec3 match-any

match ip prec 3

match mpls exp 3

then the policy map

policy-map shape_out

class exp3_ipprec3

shape average 2000000

Hope to help

Giuseppe

Chuan Liu · ‎05-25-2009

Hi Giuseppe,

Thank you very much for your reponse.

Since this is an MPLS enabled link between 2 PE routers that are connected by metraethernet via a service provider, all traffic except OSPF and BGP are MPLS labeled. In the policy-map, I have a default class defined which should catch everything:

--------

policy-map IP2MPLS-4-2-4

class QOS-GROUP5-50

priority percent 40

set mpls experimental topmost 5

class QOS-GROUP3-30

bandwidth percent 20

random-detect

set mpls experimental topmost 3

class class-default

bandwidth percent 40

set mpls experimental topmost 0

--------------------

The policy is used as a child policy so that it can be applied to the sub-interface:

--------------

policy-map IP2MPLS-Shape

class class-default

shape average 2048000

service-policy IP2MPLS-4-2-4

--------------

I can see drops in the default-class in the child policy-map. No drops have ever been seen in the priority class 'QOS-GROUP5-50', but occassionally we get choppy voice and weak voice. I suspect that the SP drop the over-subscribed traffic that includes voip traffic. This over-subscribed traffic might be the 'mpls layer' traffic.

This reasoning holds water for another similar 10M link. This is only 30% utilized when the voice is choppy. However the 'mpls layer' entry in the monitoring tool shows 90% utilization spikes. It seems to me that the 'mpls layer' entry is more accurate.

My issuse is how to take that 'mpls layer' traffic into QoS classification? Why the default class does not catch this?

Any ideas?

Thanks.

Larry

Joseph W. Doherty · ‎05-27-2009

Unable to comment on the MPLS layer stats, but in general, most monitoring isn't sufficiently granular to "see" microbursts that can impact VoIP quality.

To guarantee VoIP quality, you need to insure its packets go first when there's congestion, but this also entails insuring something like LLQ does "see" the congestion. For instance, you may need to override the default hardware FIFO queue. Or, with a shaper, use a smaller Tc then the defualt. (You also should account for vendor real-time policer values although often difficult to get them to document.) Also with shapers, you might need to account for Ethernet framing overhead. Since you mention using a subinterface, another issue is subinterfaces logically oversubscribing the physical bandwidth.