Traffic shaping
For you to achieve the best performance from the Ethernet Access product, your CPE hardware should be configured to match the purchased rate using traffic shaping. CPE traffic shaping should be implemented on all L2 or L3 CPE devices directly connecting to a Telstra NTU.
Ethernet can sometimes be confused with Carrier Ethernet, which is more stringently specified. The Ethernet Access Product is compliant with selected Metro Ethernet Forum (MEF) Carrier Ethernet specifications as set out in the product fact sheets, and certifications as indicated on the MEF Website Certification Registry.
The Ethernet Access product does not operate quite the same way as Business Grade Ethernet and Carrier Grade Ethernet products in terms of CPE configuration, so you’ll need to understand the following technical guidelines and recommendations before determining your Ethernet Access traffic shaping configuration.
Carrier Ethernet is a standardised service delivered over a wide area network. This standardised service defines:
With Ethernet Access the parameters above are configured on the Telstra Network Termination Unit provided with each service. Alignment of both Telstra NTU and Customer CPE to the Bandwidth profile values is crucial to achieve best performance The L3 CPE configuration needs to take into account the L2 service Ethernet overheads. On fibre-access UNIs we suggest allowing 2 per cent for the L2 overhead.
Therefore, for a 20M service connecting to a CPE router (L3), the service rate is calculated: 20 * 0.98 = 19.6Mbps
Configuring CPE traffic shaping
There are many different vendor implementations covering traffic shaping/policing and even differences within a single vendor’s various implementations in hardware and/or software.
The following information is provided only for configuration guidance purposes. It uses some common Cisco configurations that are representative or typical of other vendor implementations. This configuration guidance is not to be considered as an implicit vendor endorsement or recommendation.
Basic CPE configuration is to define a policy-map then apply that policy-map to the egress interface or sub-interface (VLAN).
Step 1 - Create Policy Map: policy-map <policy-name> class class-default shape average <rate in bits> <sustained burst in bits> <excess burst in bits>
- If the CPE does not accept ”0” for excess-burst, input a small value. E.g. 1000
- If you leave out the sustained & excess burst values, Cisco devices will calculate default values based on shaping rate which will result in degraded throughput.
Step 2 - Apply Policy Map to interface/sub-interface: interface ethernet<x/y.z> service-policy output <policy-name>
In the appendix of this document, you’ll find the latest shaping tables (for 1596 byte MTU) with burst values to use for Standard, Premium, Priority and Expedited CoS levels.
Values above 300Mb/s have been extrapolated and are currently undergoing validation, but have been included as they will provide better results than using the burst values in issue 1.0 of the TSIS. The TSIS is currently being revised to version 2.0 which will include the fully-validated updated tables (plus other changes throughout the document which are required for clarification purposes).
Egress shaping (at both EVC end-points) will optimise throughput. Where egress shaping is not possible, but egress policing is possible, policing is the next best option but only on Expedited COS Services, where the traffic rate is known in advance (ie Constant bit rate, such as VoIP) and the traffic uses UDP as the protocol.
On Standard, Premium and Priority, where traffic contains a high proportion of TCP, shaping is the only option to mitigate undue discards (ie reducing throughput) on ingress at the Telstra UNI. If your CPE is limited in its ability to adequately shape as described, you (or your end customer) may need to consider replacing the CPE with a device that has sufficiently granular shaping capability.
Finally, you’ll find complementary recommendations in section 9.3 of the TSIS about traffic shaping and the steps you’ll need to take to avoid excessive discards at the UNI. That section references a detailed discussion on considerations for Ethernet throughput on the MEF website here.
Appendix
