Frame Relay Traffic Shaping

Introduction

Traffic Shaping is a mechanism for controlling the rate of outgoing traffic in order to minimizes network packet loss. It delays excess PVC traffic by queuing it when traffic throughput is higher than expected and matches its transmission to the speed of the remote, target interface. Traffic shaping also avoids overloading a remote link by smoothing the traffic and regulating the average rate on an outgoing interface.

Feature Summary

Data Traffic Shaping features:
• Outgoing data traffic rate control in a range between the configured CIR an the line access rate
• Seamless transition from Traffic Shaping mode to configured Congestion Control Mode when the network is congested
• Support for existing Voice Bandwidth Allocation mechanism

Traffic Shaping -Rate Control

A configuration parameter Maximum Information Rate (MIR) is available to control the station outgoing information rate and provide traffic shaping capabilities. When the Frame Relay station parameter Congestion Control Mode is configured as NORMAL, DISABLE or LIMIT you have the option to limit the station outgoing information rate to some pre-determined value, Maximum Information Rate (MIR). The value of MIR shall be equal to or greater then CIR and equal to or less than the local interface access rate. When this option is disabled the station transmits (without rate control) at maximum line speed.

Traffic Shaping - Measurement Interval

When the Traffic Shaping feature is enabled, Measurement Interval Tc is forced to be in range of 50 to 200ms. When calculated value Tc = Bc/CIR is greater than 200ms, Tc is set to 200ms and the burst size Bmir (Bmir=Tc*MIR) will be calculated accordingly. The number of data bits, transmitted per Tc will be always kept equal to or less then Bmir.
Data throughput should not be affected by value of Tc. When Bmir is small, it is likely that just few packets can be transmitted per Tc and that some significant portion of the bandwidth may be unused. Since you are not allowed to send more than Bmir bits per Tc, the effective bandwidth is lesser than MIR. When voice is present, this effect is minimized due to the small size of voice packets as well as segmentation of data packets. When the variation in packet size is small, MIR shall be based on that size. For instance, when segmentation is enabled and End-to-End Segment Size When Voice Is Not Present is different then Disabled the MIR shall be selected to allow integer multiple of segment (including all headers) size per Tc. In such a way, when voice is not present, the data throughput will be closest to the selected MIR value. When the size of some packet is bigger than allowed Bmir, a big packet is allowed to pass by crediting in advance, to avoid packet flow blockade. When the size of the packet is bigger then Bmir it will be transmitted more than Bmir data bits per Tc, but the multi interval average rate will be kept no more than the MIR.

Traffic Shaping in Presence of Voice

Traffic Shaping applies to data traffic only. Voice packets will not be delayed or dropped because of Traffic Shaping. When there is voice traffic through PVC, a decision must be made if data throughput or voice quality is the priority. To avoid voice degradation, due to traffic shaping, the Voice Congestion Control Mode parameter shall be enabled. When both Traffic Shaping and Voice Congestion Control Mode parameters are enabled and voice is present, Voice Congestion
Control algorithm will apply (there is no traffic shaping for voice). Required bandwidth is reserved for voice on that PVC and the remaining bandwidth is used for data. At instances when voice is not present, data traffic is shaped. The data rate is then controlled by shaping the mechanism up to MIR, regardless of the Voice Congestion Control status. When Voice Congestion Control Mode is disabled, voice and data will share bandwidth determined by MIR. In all cases, when there are more bits to be transmitted than allowed, data is queued and voice is transmitted.

Traffic Shaping Typical Example

Network congestion is formally defined as “Traffic in excess of network capacity”. Consequences of congestion are long delays, as result of packet queuing in network switches and packet loss, as result of buffer overflow. Source and destination line speed mismatch in presence of bursty traffic is a frequent cause of network congestion. The central site (Node A) has a T1 line into the cloud, while the remote (branch or telecommuter) site has a lower speed (56 Kbps). The central site sends packets at a much higher rate than the remote line can transmit to the destination node (the ingress rate is higher than the egress one). This results in a bottleneck (packet queuing and eventually dropping) in the egress switch. A possible solution is rate limiting at the central site, so you do not exceed the remote side. Configuring the MIR for the corresponding stations in the node A, to 56k, limit their outgoing rates and prevent packet loss.
Traffic Shaping is a mechanism for controlling the rate of outgoing traffic in order to minimizes network packet loss. It delays excess PVC traffic by queuing it when traffic throughput is higher than expected and matches its transmission to the speed of the remote, target interface. Traffic shaping also avoids overloading a remote link by smoothing the traffic and regulating the average rate on an outgoing interface.

Frame Relay Transmission Fairness

Introduction
The transmission of frames is regulated on each DLCI so that one DLCI carrying intense traffic and/or large frames, does not effect other DLCIs on the same link. Effectively, this shares the link transmission bandwidth amongst all DLCIs and ensures that, at the very least, each DLCI obtains its CIR. Transmission Fairness is beneficial to those stations that are constantly transmitting. Stations that transmit at infrequent intervals typically operate well below their CIR and, as such, transmit their data when necessary.
Transmission fairness does not imply a priority level. When the occasional frame is queued for transmission by a station, the frame waits its turn in the queue. Having a large CIR, with respect to other stations, does not mean that the frame is moved up in the queue. This would however, apply to voice packets.
Since no priority is associated with transmission fairness, there is no overall performance change when using pre-5.1 release software. This is especially true of applications using a low window number for its interworking with a remote. A typical example of this is an application that sends a single message and waits for an acknowledgment before sending the next message.

Sharing Link Bandwidth
The sharing of link bandwidth is in proportion to the stations CIR. The amount of additional bandwidth given a station is determined by its configured CIR, and the sum of CIRs from all transmitting stations.
When uncommitted bandwidth is available, it is divided between all transmitting DLCIs.

Zero CIR Configuration
When zero-CIR stations are configured, the Clock Speed parameter must be set to a value representing the actual link operating speed. This also applies to ports that are externally clocked.
Since all DLCIs with a non-zero CIR may periodically saturate the link, DLCIs with a zero CIR are not guaranteed bandwidth on the link.
These two formulas apply to non-congested conditions (when the station is not in a controlled sending state). For the purposes of the calculations, only actively transmitting stations are considered. Nt denotes the total number of such stations while the number of such stations with CIR set to zero.
• The fraction of link bandwidth available to each zero CIR station (Fz) is
calculated here
• The fraction of link bandwidth available to each non-zero CIR station (Fn) is
calculated here:
Fz = 1- (total CIR / link speed)
Nt
Fn = Station CIR
total CIR * (1 - Fz * )

Example
For the purposes of this example, assume that a node has the following conditions:
• FRI port has a link speed of 64 kbps.
• All stations have a CIR of 16 kbps.
• Three Bypass stations carrying LAN traffic (stations 1, 2, and 3) and one
Annex G station carrying serial traffic (station 4).
Station 1 is idle.
Stations 2 through 4 are actively transmitting data.
Since each station is configured with the same CIR, the amount of bandwidth given
each active station is the same:
(16 x 64)/(16+16+16) = 21.3 kbps