Traffic and Resource Management

The traffic management features of the ATM switch router provide the following capabilities:

• Integrated support for different types of ATM services

• Flexible policies for bandwidth allocation through controlled link-sharing services

• Effective use of network resources

The congestion control capabilities of the ATM switch router support the following goals:

• Avoid conditions where congestion can occur

• Minimize the effects of congestion when it does occur

• Prevent spreading the congestion state to other parts of the network

Because ATM networks are designed to carry many different types of traffic, traffic characteristics and QoS requirements of each virtual connection must be described, and delivery of the contract must be guaranteed within the resource allocation policies defined for the network.

These requirements are carried out in three phases:

1. Define the traffic and service contract.

2. Find an acceptable path for the connections.

3. Use hardware resources to honor the terms of the contract for the life of the connection.

The first of these two steps can be considered the connection setup phase, while the third step represents the data flow phase. These three phases and their supporting mechanisms are discussed in the following sections.

• A connection traffic descriptor, which includes a service category and applicable parameters

• An optional set of QoS parameters applicable to the service category

When establishing the traffic and service contract, target values for QoS parameters can be used as criteria for the connection setup requirements. These values are either metrics (accumulated over multiple hops of a call) or attributes (a gating criterion that is not accumulated, but is checked at each interface). Maximum cell transfer delay (maxCTD) and peak-to-peak cell delay variation (ppCDV) are metrics, while cell loss ratio is an attribute.

• The connection traffic table rows

• The default QoS objective table

• The default MBS

• The default CDVT

Configuration Overview

Configuring the default QoS objective table requires one or more of the following steps; each objective can have a defined or undefined value.

• Specify a maxCTD value in microseconds for CBR, VBR-RT, or for both.

• Specify a ppCDV value in microseconds for CBR, VBR-RT, or for both.

• Specify a cell loss ratio exponent for CBR, VBR-RT, VBR-NRT, or for all three.

You can specify separate loss ratio exponents for CLP0 and CLP0+1 cells.

The default QoS objective table should be configured with the same values for an entire network.

Resource CAC (RCAC) uses the following information provided in the traffic contract for each direction of a requested connection:

1. Parameter values of the source traffic descriptor---PCR, SCR, MCR, MBS, CDVT

2. The requested service category (the service category must be the same for both directions of a connection) or QoS parameters (CLR, CTD, CDV) or both service category and QoS parameters

RCAC is based on a proprietary Equivalent Bandwidth algorithm in which equivalent bandwidths are used as real constant bandwidths in a statistical time-division multiplexing (STM) CAC environment for fast calculation. The equivalent bandwidth (or effective bandwidth by some) of a source is defined to be the minimum capacity required to serve the traffic source so as to achieve a specified steady-state CLR, maxCTD, and ppCDV for CBR/VBR and for the nonzero MCR portion of ABR/UBR+ connections.

• SCRM (SCR margin)---extra equivalent bandwidth (over and above SCR) allocated to a VBR connection to account for bursting to PCR.

• SCRMF (SCRM factor)---configurable safety margin with default = 1%

• R_CLR_CBR:CLR---estimate for CBR connection transiting an interface

• R_CLR_VBR:CLR---estimate for VBR connection transiting an interface

• R_ppCDV_CBR---estimate on maximum transmit ppCDV that might be experienced by a CBR connection transiting an interface

• R_ppCDV_VBR---estimate on maximum transmit ppCDV that might be experienced by a VBR-RT connection transiting an interface

• R_MaxCTD_CBR---estimate on maximum transmit maxCTD that may be experienced by a CBR connection transiting an interface

• R_MaxCTD_VBR---estimate on maximum transmit maxCTD that may be experienced by a VBR-NRT connection transiting an interface

• MAXCR---Maximum cell rate in a direction on an interface

• MAX_BE_CONNS---user-configured maximum number of best-effort (ABR/UBR) connections allowed on an interface

• MAX_PCR_CBR, MAX_PCR_VBR, MAX_PCR_ABR, MAX_PCR_UBR---user-configured maximum allowed PCR, per service category (default = line rate)

• MAX_SCR---user-configured maximum allowed SCR (default = line rate)

• MAX_MCR_ABR, MAX_MCR_ABR--- user-configured maximum allowed MCR for ABR, UBR+ (default = line rate)

• MAX_MBS---user-configured maximum allowed MBS (default = none)

• MAX_CDVT_CBR, MAX_CDVT_VBR, MAX_CDVT_ABR, MAX_CDVT_UBR:---
  user-configured maximum allowed CDVT, per service category (default = no limit)

Parameters used in the algorithm are defined as follows:

• CBR_EQ_BW, VBR_EQ_BW, ABR_EQ_BW, UBR_EQ_BW---the equivalent bandwidth proposed for a CBR, VBR, ABR, or UBR+ connection.

Input:
Traffic contract input:   service_category, PCR, SCR, SCRMF, MBS, CDVT, MCR
			   tx_MaxCTD_obj, tx_MaxCTD_acc, tx_ppCDV_obj, tx_ppCDV_acc, CLR 
 
output:CAC_accept - true,  connection is accepted
                     false, connection is rejected 
 
algorithm:
 
Based on service category of the proposed connection:
 
service_category == CBR:
 
   if ((CLR >= R_CLR_CBR) && 
       (tx_MaxCTD_obj >= R_MaxCTD_CBR + tx_MaxCTD_acc) && 
       (tx_ppCDV_obj  >= R_ppCDV_CBR + tx_ppCDV_acc) && 
       (PCR <= MAXCR) && 
       (PCR <= MAX_PCR_CBR) &&
       (CDVT <= MAX_CDVT_CBR) {
 
          CBR_EQ_BW = PCR
          if (CBR_EQ_BW >  ACR_CBR) 
             CAC_accept = false 
          else 
             CAC_accept = true  
   } else
      CAC_accept = false 
 
service_category == VBR-RT:
 
   if ((CLR >= R_CLR_VBR) && 
       (tx_MaxCTD_obj >= R_MaxCTD_VBR + tx_MaxCTD_acc) &&
       (tx_ppCDV_obj  >= R_ppCDV_VBR  + tx_ppCDV_acc) && 
       (PCR <= MAXCR) && 
       (PCR <= MAX_PCR_VBR) && 
       (SCR <= MAX_CR) && 
       (SCR <= MAX_SCR) && 
       (MBS <= MAX_MBS) &&
       (CDVT <= MAX_CDVT_VBR) {
 
          SCRM = SCRMF * (PCR - SCR) /* SCRMF = [0,...,1] w. default=0.01 */
          VBR_BW = SCR + SCRM
          if (VBR_BW >  ACR_VBR)
             CAC_accept = false
          else 
             CAC_accept = true 
   } else
      CAC_accept = false 
 
service_category == VBR-NRT:
 
   if ((CLR >= R_CLR_VBR) &&
       (PCR <= MAXCR) &&
       (PCR <= MAX_PCR_VBR) &&
       (SCR <= MAX_CR) &&
       (SCR <= MAX_SCR) &&
       (MBS <= MAX_MBS) &&
       (CDVT <= MAX_CDVT_VBR) {
 
          SCRM = SCRMF * (PCR - SCR) /* SCRMF = [0,...,1] w. default=0.01 */
          VBR_BW = SCR + SCRM
          if (VBR_BW >  ACR_VBR)
             CAC_accept = false
          else 
             CAC_accept = true 
   } else
      CAC_accept = false 
 
service_category == ABR:
    
   if ((PCR <= MAXCR) &&
       (PCR <= MAX_PCR_ABR) &&
       (MCR <= MAXCR) &&
       (MCR <= MAX_MCR_ABR) &&
       (CDVT <= MAX_CDVT_ABR) &&
       (ABR_count + UBR_count + 1  <= MAX_BE_CONNS)){
 
          CBR_EQ_BW = PCR
          if (CBR_EQ_BW >  ACR_CBR) 
             CAC_accept = false 
          else 
             CAC_accept = true  
   } else
      CAC_accept = false 
 
service_category == UBR:
 
   if ((PCR <= MAXCR) &&
       (PCR <= MAX_PCR_UBR) &&
       (MCR <= MAXCR) &&
       (MCR <= MAX_MCR_UBR) &&
       (CDVT <= MAX_CDVT_UBR) &&
       (ABR_count + UBR_count + 1  <= MAX_BE_CONNS)){
 
          CBR_EQ_BW = PCR
          if (CBR_EQ_BW >  ACR_CBR)
             CAC_accept = false
          else
             CAC_accept = true
   } else
      CAC_accept = false
 

Note that the above algorithm does not describe the derivation of available cell rate (ACR) per service category. In the absence of controlled link sharing, this algorithm applies to each direction:

(.95 * MAXCR - sum of equivalent bw allocated to all connections on interface)
 

If controlled link sharing is configured, it then establishes limits on ACR for all guaranteed bandwidth or a service type (the maximum case) and limits on the encroachment of other service types (the minimum case).

• Sustained cell rate margin factor

• Controlled link sharing

• Outbound link distance

• Limits of best-effort connections

• Interface maximum of individual traffic parameters

• Service category support per interface

bandwidth = (SCRMF * (PCR-SCR))/100 + SCR

Changing the outbound link distance from its default of zero can affect the switched virtual connection requests accepted. For example, you might want to discourage use of a transcontinental link by configuring a higher propagation delay.

The default for physical interfaces and hierarchical VP tunnels is to allow virtual connections of any service category to transit the interface. However, interface service category configuration can be used explicitly to allow or prevent virtual connections of specified service categories to migrate across the interface.

• Policing ---Usage Parameter Control (UPC)

• Cell queuing

• Congestion notification

• Output scheduling

For most applications, when one or more cells are dropped by the network, the corresponding packet becomes corrupted and useless. This results in the need to retransmit the many cells that comprise that packet, and leads to exacerbated congestion. For example, loss of a cell from an IP over ATM packet (RFC 1577) might require resending 192 ATM cells, given an MTU of 9 KB.

To maximize the number of complete delivered packets, the ATM switch router implements a unique tail packet discard and early packet discard (TPD/EPD) scheme that intelligently and selectively discards cells belonging to the same packet. These congestion control mechanisms reduce the effects of fragmentation and make the ATM switch router essentially emulate a packet switch, which discards entire packets.

UPC on the ATM switch router checks the following parameters:

• PCR and CDVT for CBR, UBR, and ABR connections

• SCR and MBS for VBR connections

On systems equipped with hardware to support the dual leaky bucket, there are two policers for VBR connections. One policer uses PCR and CDVT, while the other policer uses SCR and MBS.

Configuration Overview

Configuring the default UPC behavior requires the following steps:

Step 1 For permanent virtual connections, specify UPC tag or drop in the command when you configure the PVC or PVP; the behavior for each half-leg of the connection can be separately configured.

Step 2 For SVCs, select an interface, enter interface configuration mode, and specify tag or drop.

You can specify CDVT or MBS for PVCs through a connection traffic table row. If no CDVT or MBS is specified in the row, then a per-interface, per-service category default is applied for purposes of UPC on the connection.

Configuration Overview

Configuring the default CDVT and MBS on an interface requires the following steps:

Step 1 Select the interface to configure and enter interface configuration mode.

Step 2 Specify a CDVT default value for a service category. You can repeat this step for additional service categories you want to configure.

Step 3 Specify an MBS default value for a service category. You can repeat this step for additional service categories you want to configure.

The specific features available depend upon the hardware:

• For systems with the FC-PCQ

  • Oversubscription factor

  • Service category limit

  • Maximum queue size per interface

  • Interface queue thresholds per service category

• For systems with the FC-PFQ and the Catalyst 8540 MSR, threshold groups can be configured.

The size of the VBR-NRT queue and ABR/UBR queues is determined by using the oversubscription factor (OSF) in the following formula:

size (vbr-nrt) = .25 * ((osf * 2048) - DefaultSize (cbr) - DefaultSize (vbr-rt))
size (abr-ubr) = .75 * ((osf * 2048) - DefaultSize (cbr) - DefaultSize (vbr-rt))
 

When you configure the oversubscription factor, you are changing the default values. Refer to the ATM Switch Router Command Reference publication for details.

Configuration Overview

Configuring the oversubscription factor requires the following steps:

Step 1 From global configuration mode, specify a value for the oversubscription factor.

Step 2 To have the change take effect and resize the queues, save the running configuration to the startup configuration and restart the ATM switch router.

When you configure the service category limit requirements, you are specifying a new value to use rather than the default. To do so requires just one global configuration command in which you specify the limit in number of cells for a queue type. You repeat this command for each additional queue type for which you want to configure a maximum.

Configuration Overview

Configuring the maximum queue size for an interface requires the following steps:

Step 1 Select the interface to configure and enter interface configuration mode.

Step 2 Specify an output queue maximum size for a service category. Repeat for additional service categories you want to configure.

The following queue thresholds can be configured per interface queue:

• Threshold at which the EFCI bit is set

• Threshold at which cells are eligible for CLP discard or EPD

• Threshold for relative rate (RR) marking on ABR connections

Configuration Overview

Configuring the interface queue threshold per service category requires the following steps:

Step 1 Select the interface to configure and enter interface configuration mode.

Step 2 Specify an output percent for the EFCI marking threshold for a service category. You can repeat this step for additional service categories on the interface.

Step 3 Specify an output percent for the discard threshold for a service category. You can repeat this step for additional service categories on the interface.

Step 4 Specify an output threshold percent for RR marking on ABR connections.

Each threshold group has a set of eight regions, and each region has a set of thresholds. When these thresholds are exceeded, cells are dropped to maintain the integrity of the shared memory resource.

The initial default configuration of per-VC queuing on the ATM switch router has all the connections of a service category assigned to one threshold group. However, the assignment of service categories to threshold groups is configurable, with the following restrictions:

• A service category cannot be mapped to more than one threshold group.

• If you configure a service category to a threshold group more than once, only the last configuration is in effect.

• The default assigns each service category to a different threshold group. However, you can assign more than one service category to a threshold group.

The threshold group configuration parameters are as follows:

How It Works

When congestion is in the range of 0 cells (uncongested) to 1/8th full, the connection queues are limited to max-queue-size. When congestion is in the range of 7/8ths full to completely full, the connection queues are limited to min-queue-size.

Configuration Overview

Configuring the threshold group parameters requires the following steps:

Step 1 From global configuration mode, assign a service category to a threshold group (1-6).

Step 2 Specify the maximum number of cells queued for all connections that are members of the threshold group.

Step 3 Specify the percent at which the per-connection queue is to be considered full for purposes of CLP discard and EPD.

Step 4 Specify the maximum per-connection queue limit (in number of cells) for the threshold group.

Step 5 Specify the minimum per-connection queue limit (in number of cells) for the threshold group.

Step 6 Specify a name to associate with the threshold group (optional).

Step 7 Specify the percent at which the per-connection queue is considered full for EFCI (on all connections) and RR marking (on ABR connections).

---

Note RR marking is not supported on all platforms.

---

You can repeat these steps for any additional service category thresholds you want to configure.

• EFCI marking (for non-ABR connections)---In this mode the ATM switch router sets the EFCI state in the headers of forward data cells to indicate congestion. When the destination receives an EFCI flag, it marks the congestion indication bit in the resource management cells to indicate congestion and sends the resource management cells back to the destination.

• Relative rate (RR) marking mode (for ABR connections)---In this mode an ATM switch router that experiences congestion can set the congestion indication bit directly in forward and backward resource management cells. RR marking mode is used only for ABR connections and is not supported on all hardware platforms.

EFCI and RR marking involve two important functions: detecting incipient congestion and providing selective feedback to the source. As with any feedback mechanism, congestion control schemes operate best when the latency of the feedback path is minimized. Thus RR mode, because of its ability to use backward RM cells to send the congestion indicator rather than relying on the destination end system to reflect it back, can greatly reduce feedback delays and deliver better performance than the EFCI mode.

The two modes can be used independently or in combination to support ABR traffic, and thresholds can be set to indicate when EFCI or RR marking should occur.

Configuration Overview

When you configure the ABR congestion notification mode you affect all ABR connections. To do so requires just one global configuration command.

Configuration Overview

Interface output pacing is disabled by default on the ATM switch router. Configuring the interface output pacing requires the following steps:

Step 1 Select the interface to configure and enter interface configuration mode.

---

Note There are restrictions on which interface types can be configured for output pacing; refer to the ATM Switch Router Command Reference publication for details.

---

Step 2 Specify the output pacing limit as a a bit rate in Kbps.

This configuration does not take effect if the amount of bandwidth allocated to CBR and VBR connections in the transmit direction on the interface is greater than the configured output pacing value.

In scheduling the next cell to be transmitted from a port, RS has first call on supplying an eligible cell. If RS does not have one, then the WRR scheduler chooses a service class with an output virtual connection ready to transmit, and finally a virtual connection within the service class is selected.

If the timewheel RS does not have an output virtual circuit ready to transmit, the WRR scheduler becomes active to pick out a virtual circuit to transmit a cell. The WRR scheduler uses the interface bandwidth left over after guaranteed cell service to transmit cells. Thus, an output virtual circuit of a service category other than CBR can be serviced by both the rate scheduler and the WRR scheduler. A CBR output virtual circuit cannot be serviced by the WRR scheduler because its PCR is already guaranteed by the rate scheduler. (Any additional cell transmission by the WRR scheduler out of that output virtual circuit is likely to arrive too soon at the next switch and might be policed.)

The following service categories can be serviced by the WRR scheduler:

• VBR-RT

• VBR-NRT

• ABR

• UBR

Each service class is assigned a weight. These weights are configurable, in the range of 1 to 15. The default weighting is {15,2,2,2} for classes {2,3,4,5}, respectively. The weighting is not modified dynamically.

Within service classes, individual output virtual circuits are also weighted, again in the range of 1 to 15. A standard weight (2) is assigned to all PVCs in a service class. Optionally, permanent virtual circuits can be configured with a specific weight per half-leg (applying to the transmit output virtual circuit weight). Switched virtual circuits take the value 2.

Configuration Overview

Configuring the service class weights on an interface requires the following steps:

Step 1 Select the interface to configure and enter interface configuration mode.

Step 2 Specify a service class and a weight value. You can repeat this step for additional service classes.