Trickling Egress Frame Discards and PIF Overflows [Cisco WAN Switching Modules]

Document ID: 10797

Updated: Oct 04, 2005

Download PDF

Details

Introduction

When Packet Input FIFO (PIF) overflows occur because of excessively high credit maximum (Cmax) settings on Cisco Internetwork Packet Exchange (IPX) and IGX platforms, isolation and diagnosis can be difficult. Trickling discards in conjunction with high Cmax settings and large port rates are typical signs of an IPX or IGX network with PIF overflow problems.

These PIF overflows are the result of an inability of the FIFO buffer to meet the instantaneous demand to process packets. Overflows that high Cmax settings cause can occur in networks with both ForeSight and non-ForeSight permanent virtual circuits (PVCs). Networks that use ForeSight in combination with high Cmax and high port rate settings are especially vulnerable to PIF overflows. In addition, this combination can hinder the ability of ForeSight to effectively manage PVC rates. To avoid these complications, use a Cmax setting of 10.

Prerequisites

Requirements

There are no specific requirements for this document.

Components Used

This document is not restricted to specific software and hardware versions.

Conventions

For more information on document conventions, refer to the Cisco Technical Tips Conventions.

Explanation

Terminology

Trickling egress errors is a term that refers to a low rate of consistent errors or frame discards. The cause of these discards can be difficult to determine because many possibilities exist.

Cmax is the number of packets a connection can burst into the network at port speed without a throttle to minimum information rates (MIR) or quiescent information rates (QIR). Generally, configure Cmax to approximate the size of the average frame you expect from user equipment.

ForeSight is a proprietary, dynamic closed-loop, rate-based, congestion management feature that yields bandwidth savings at the transmit of bursty data across cell-based networks. ForeSight ensures that connections get the maximum available throughput and that no connection takes the capacity that other connections require.

Details

Trickling egress frame discards can occur on connections with high Cmax values and high port rates. Cmax is a configurable parameter for Frame Relay fixed rate and ForeSight connections. The Cmax value is the maximum number of credits a Frame Relay connection accrues during idle time. Each credit equals one FastPacket.

These credits allow connections to burst a set quantity of FastPackets, with no interruption, from the IPX or IGX Frame Relay card. When a connection is able to burst with use of its credits, there is no mechanism, including ForeSight, that can throttle the burst of packets regardless of congestion along the PVC path. Therefore, on Frame Relay networks that use ForeSight, a Cmax value of 10 gives the best overall performance. Though this parameter is configurable, do not change the default setting to a value greater than 10.

If you set the Cmax value above 10, you can obstruct the effective management by ForeSight of the network PVCs. In addition, packet discards can occur on trunks, even during times of moderate trunk congestion. Networks with high Cmax values in conjunction with large port rates have a higher probability of packet discards because of PIF overflows. Trickling egress discards are typical symptoms of PIF overflows. Sometimes, you can attribute trickling discards to ingress discards or trunk errors. But, in the case of PIF overflows, there is no obvious explanation for the occurrence.

To identify PIF overflows, you must perform Cbus traces on all through and terminating nodes for a specific PVC. The PIF on the trunk card can hold 64 FastPackets and has an approximate throughput rate of 4 Mbps. When overflows occur, the FIFO buffer is unable to pass the packets to the trunk rapidly enough to meet the instantaneous demand of packets input from the bus.

This event can occur so rapidly that, when you use the dsptrkutl command to average the event over the one-second update, the event does not have a significant impact on the statistics that display.