cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
1050
Views
0
Helpful
3
Replies

Tc calculation for GE and FE handoff to SP for VoIP

Michael Grann
Level 1
Level 1

Hi,

I have a question pertaining to which option is preferred between the choice of a GE and a FE connection to the SP MPLS cloud. If you are running VoIP and subscribed to EF service, and your CIR is for example 90Mbps, are you better off provisioning the link with FE instead of GE? The reason being that since GE's clockrate is so much faster than FE, that would mean a much shorter Tc interval to serialize VoIP out the interface, and thus longer "wait time" for the next Tc instance when calculating the rate of refilling the buckets with tokens. Is this correct and does this have any practical implications? Do you have any link on CCO that compares the two?

TIA.

MG

1 Accepted Solution

Accepted Solutions

MG,

Tc = Bc / CIR

AFAIK, the Tc default depend on each platform.

If CIR and Bc are equal in your GE and FE policies, Tc would be equal as well.

If you want to, you can set the Tc you need configuring appropriate CIR and Bc values in the police command (MQC), if the Bc min and max values allow you to do so.

The policer granularity may not allow you to get an exact number, but you should be able to get close to it.

HTH, Gustavo

View solution in original post

3 Replies 3

Andre Gustavo Albuquerque
Cisco Employee
Cisco Employee

"serialization delay" is one of the components to be used when calculating the end-to-end delay of your time sensitive service.

You'd also have to add: coding/decoding delay, propagation delay, switching delay, and scheduling delay.

It would be good to first determine what are the correct service level requirements for this type of service, and several factors should be considered for VoIP, such as the Codec which is being used and also the intended quality of the service being delivered.

Having defined these parameters, you should be able to determine yourself whether a FE is enough.

To give you a perspective about this diference, see the example below:

- Codec G.711 (64kbps), sample interval 10ms, payload 80 bytes

- L3 Overhead: 40 bytes = 20 bytes (IP) + 8 bytes (UDP) + 12 bytes (RTP)

- L2 Overhead: 18 bytes (Ethernet L2 headers, including 4 bytes of FCS or CRC)

=> total packet size: 138 bytes

The serialization delay of this packet in a FE link is 11ms and in a GE link is 1ms.

There are some studies relating latency and perceived voice quality, per codec being used. There are some models used for that, such as R factor and MOS (Mean Opinion Score). Based on that, you can determine your end-to-end delay budget.

I recommend reading the book "Deploying IP and MPLS QoS for Multiservice Networks", by John Evans and Clarence Filsfils.

HTH, Gustavo

Hi Andre,

Thank you for the explanation and the reference to the book. Let me rephrase my question - with all parameters affecting delay being equal, would the fact that a GE's clock rate and serialization being 1/10th that of the FE have any negative impact on the Tc interval's calculation of how frequently the token buckets are calculated? I remember something in either Wendel Odom's QoS book or the AQOS class that mentions keeping the WAN interface handoff as close to the CIR as possible (thus, it would be a better choice to have FE with a 75Mb CIR as opposed to using GE).

Thanks again.

MG

MG,

Tc = Bc / CIR

AFAIK, the Tc default depend on each platform.

If CIR and Bc are equal in your GE and FE policies, Tc would be equal as well.

If you want to, you can set the Tc you need configuring appropriate CIR and Bc values in the police command (MQC), if the Bc min and max values allow you to do so.

The policer granularity may not allow you to get an exact number, but you should be able to get close to it.

HTH, Gustavo