Login to your account

When enforcing port-based traffic shaping to throttle back a Gigabit per second to say 8 mbps, where should you do it

a) at the Network-to-user Handoff on the NID RJ45 port ?
b) at the NID-to-Network on the NID SFP port ?
c) at the Network-provider switch on the SFP port facing the NID SFP port ?
d) at a,b & c ?
e) b,c ?

I seem to be able to do it only at the A) location above, and it seems to work. However, I am worried that this taxes the resources of the NID (under a DDOS scenario for instance).

What's your opinion ?

F.

Corollary question:

MEF equipment provisioned for E-Line services by a provisioning tool, would configure both the NID as well as its host switch, to do the traffic shaping on both the NID 'wan' port, as well as the host switch facing the NID ?

F.

Francois,

The Bandwidth profile is what people refer to as "policing" which is bufferless measurement of the offered traffic.
MEF recommends shaping before this point.
The subscriber's network element (be there host, switch or router) is referred to as the CE. It is not managed by the service provider.
Assuming that the subscriber purchased E-Line with 8 Mbps CIR and some CBS, it is advisable that the CE would shape the traffic towards the WAN to 8 Mbps.
If it can't, then the NID might be doing it for him. This assumes that the bandwidth profile enforcement is done at the MEN's edge device and that the NID is willing to accept burst up to the line speed (say 1 Gbps) and shape to 8 Mbps.
For doing so, it will need of course decent buffering capabilities…if you run out of buffers and start dropping too soon, then shaping would only marginally help you.

Hope this helps.

My 2 cents,
Ayal

So are you saying that the customer is expected to shape itself before hitting the service provider managed NID such as not to overcome the buffers ? It is plausible that a cost effective NID would not be able to take a burst of 1G and choke it down to the enforced E-Line service of 8 mbps, without doing something nasty. So I get it that the customer should shape itself.

However, in the other direction, would it not be advisable that the service provider host switch in the central-offic e/POP/datacenter, also shape the port facing the NID, such as also to relieve the NID from having to potentially 1G of burst to shape down to 8 mbps ?

I'm not able to read in your post whether you are saying that the Host switch should also shape on the port facing the outside plant and the NID ultimately.

F.

Hi,

Yes indeed.
But this is not a special case for NIDs.
MEF 10.2 section 10.3 provides information on the benefits by having the CE shape the traffic before sent over the UNI interface.

As to the egress direction from the SP's network to the CE:
The total BW is conforming to CIR/EIR over some reasonable time interval (since this traffic passed ingress BWP).
Now the question is whether the CEN may introduce burstiness and whether or not it will shape the traffic over the UNI.
I think that this is implementation specific. I'm not sure that all network elements of all SPs can shape traffic per service (VLAN or group of VLANs). But if it is single service (e.g. EPL) then I'm certain that most networks can do that and indeed should.

Ayal

Ayal Lior wrote:
...
Now the question is whether the CEN may introduce burstiness and whether or not it will shape the traffic over the UNI.
I think that this is implementation specific. I'm not sure that all network elements of all SPs can shape traffic per service (VLAN or group of VLANs). But if it is single service (e.g. EPL) then I'm certain that most networks can do that and indeed should.

Ayal

Hi Ayal,

The most common approach supported by the vast majority of vendors would be 4 to 8 queues per UNI port. Some equipment is differentiated by providing hierarchical TM, however this is less common.

Best
Alex