New to MPLS

bretjaquish · ‎05-15-2009

Hey all,

I am new to MPLS and would like to see if anyone can offer some insight to my questions.

We have a OC-3 connection to our MPLS provider (AT&T). We have various (100+) remote sites all over the global that have different connections to the MPLS network (T1, T3, 10Mb ethernet, etc.).

Is there any protection inherent within MPLS that would prevent the oversubscription of smaller links (OC-3 to T1)?

Is it possible my provider might limit traffic?

How could I possibly write a QoS policy when 20% of the head-end is larger than 100% of the remote end?

I'm trying to find the "nicest" solution in regarding to TCP/IP. I hear that policing is rather harsh on the windowing mechanism, and that shaping is a better option.

Would it be as simple as creating some type of nested QoS policy?

Laurent Aubert · ‎05-15-2009

Hi,

You have two options:

1- Use your SP QoS services. In this case the you will classify your traffic into the SP classes and the shaping/prioritization will be done on the remote PE connected to your remote CE.

2- Create a class-map on your hub site per remote site and shape each class-map based on the remote site link access rate.

The main limitation of this approach are:

- you need to determine what is the equivalent bw for an OC3 link of your remote site access rate (due to different L2 encapsulation so different overhead)

If you don't do it precisely, you may never reach the congestion point (so like doing nothing) or reach it too early and not able to use all your remote bw

- This model doesn't work if you have site-to-site direct communication.

So my recommendation is to go for option 1 if you can.

HTH

Laurent.

Joseph W. Doherty · ‎05-17-2009

"Is there any protection inherent within MPLS that would prevent the oversubscription of smaller links (OC-3 to T1)? "

Within the provider, normally no.

"Is it possible my provider might limit traffic? "

Often only traffic such as DSCP EF is rate-limited, although other traffic can be out-of-contract and more likely to be dropped if there's congestion.

"How could I possibly write a QoS policy when 20% of the head-end is larger than 100% of the remote end? "

Often that's done by shaping the head-end to match the remote end's bandwidth.

"I'm trying to find the "nicest" solution in regarding to TCP/IP. I hear that policing is rather harsh on the windowing mechanism, and that shaping is a better option."

On many Cisco platforms/IOSs, default settings for both policing and shaping often do make policing more "harsh". However, both can and do drop packets, and with TCP windowing, drop management is an important but often overlooked aspect of QoS. (NB: BTW, [W]RED pursues drop managment, but just one aspect.)

"Would it be as simple as creating some type of nested QoS policy? "

It might be, but as Laurent has noted in this thread, and as other posters have noted in your other similar question you've posted in the WAN forum, often MPLS provides any-to-any. For example, consider three sites all with T1 but two attempt to send to one.

[edit]

Again (as mentioned in other thread), if your logical topology is hub-and-spoke, then you can often manage the hub end to not overrun the spokes.

bretjaquish · ‎05-18-2009

Thank you both. I think this will get me in the direction that I need to look.