Brought to you by EarthWeb
IT Library Logo

Click Here!
Click Here!


Search the site:
 
EXPERT SEARCH -----
Programming Languages
Databases
Security
Web Services
Network Services
Middleware
Components
Operating Systems
User Interfaces
Groupware & Collaboration
Content Management
Productivity Applications
Hardware
Fun & Games

EarthWeb Direct EarthWeb Direct Fatbrain Auctions Support Source Answers

EarthWeb sites
Crossnodes
Datamation
Developer.com
DICE
EarthWeb.com
EarthWeb Direct
ERP Hub
Gamelan
GoCertify.com
HTMLGoodies
Intranet Journal
IT Knowledge
IT Library
JavaGoodies
JARS
JavaScripts.com
open source IT
RoadCoders
Y2K Info

Previous Table of Contents Next


Switch Behavior

A switch may use at least one feedback mechanism to control congestion. It must be pointed out that this is not an area of standardization and that it is purely up to the switch manufacturer or service provider to decide which scheme to implement.

Binary Feedback

The binary feedback scheme is based on the EFCI bit. A switch monitors the queue length and marks the EFCI bits of passing ATM cells, as long as the queue length exceeds a predefined queue threshold value. The binary feedback scheme is known to suffer from a potential unfairness when all connections share a common queueing buffer. Typically, a connection with more hops has a better chance of running into a congested switch than those with a smaller number of hops.

Explicit Rate Feedback

The explicit rate feedback scheme can provide the fairness by having each switch determine the suitable rate for an SES not to overload the switch. This rate is sent to the SES as the explicit rate. At the same time, each SES declared its prevailing cell rate in the current cell rate (CCR) field of an RM cell.

The enhanced proportional rate control algorithm (EPRCA) computes the mean ACR (MACR) among all connections as a running exponential weighted average (MACR = (1-µ)MACR + µCCR). A typical value of μ is chosen to be 1/16 and CCR is taken from the passing RM cell. The fair share is then taken as a fraction (e.g., 7/8) of MACR, and any SES sending more than the fair share is asked to reduce its rate by setting the explicit rate field in the backward RM cell to the fair share.

Congestion Avoidance

Congestion avoidance schemes monitor the prevailing load according to the load factor z = Input_Rate / Target_Rate. The input rate is measured over a fixed averaging interval, whereas the target rate is set slightly below the ABR bandwidth.

There are two variations of this scheme that differ in the manner by which the fair share or the explicit rate is computed. In the explicit rate indication for congestion avoidance (ERICA) algorithm, the fair share is given as the target rate divided by the number of active connections. A switch determines the explicit rate as ER = max (CCR/z, Fair-Share). ER is updated periodically using current cell rate information from the forward RM cells and the load during the averaging interval.

The ERICA algorithm attempts to guarantee at least the fair share amount of capacity to each active connection. Any excess capacity is distributed to active sources in proportion to their rates.

The congestion avoidance using proportional control (CAPC) algorithm, on the other hand, does not update ER for each connection, but instead uses the same ER for every connection. ER in CAPC is computed and updated from the fair share. If z < 1, Fair-Share = Fair-Share *(1 + (1-z)*Rup). Otherwise, Fair-Share *(1 - (z-1)*Rdn). In these cases, Rup and Rdn are constant slope parameters to increase and decrease the rate (the amount of changes allowed in each update is limited as well).

Assessment of ABR Control

In general, it is difficult to draw a general comparison among a number of ABR control schemes, since source and switch behaviors are affected by many factors, including the network topology. In order to have some idea about the effectiveness of different control strategies, three control algorithms are considered:

  Binary feedback control.
  EPRCA.
  ERICA.

Three ABR sources send cells to the switch buffer. They are located at 1, 10, and 100 km from the switch. Exhibits 4-6-4, 4-6-5, and 4-6-6 plot changes in ACRs in three sources.


Exhibit 4-6-4.  ACR According to the Binary Feedback Control


Exhibit 4-6-5.  ACR Changes in EPRCA


Exhibit 4-6-6.  ACR Changes in ERICA

Clearly, the simple binary feedback algorithm shows a significant oscillation, whereas the oscillation is eliminated in ERICA. The cause of oscillation is that in a high-speed network such as an ATM network, the latency for feedback information can be significant. By the time, the source recognizes congestion in the network, a few megabytes of data may already have been sent out.

When in congestion, every source attempts to reduce its rate, producing an excess capacity a short time later. When not in congestion, every source attempts to transmit a little bit more, producing a congestion a short time later. In general, however, oscillation of cell rates does not produce an adverse effect. Oscillation simply means that a source cannot send at a constant rate for an extended period of time.

Both the latency and the rate oscillation will not be of any concern in a LAN environment where the delay and reaction time are considerably short. As soon as LANs are connected over a long-distance, however, control parameters should be tuned to account for potential oscillatory behavior.

SUMMARY

In summary, network techniques to recover from congestion and limit it, such as usage parameter control and traffic policing, are successfully applied to ATM constant bit rate and variable bit rate service. The generic cell rate algorithm, as specified by the ATM Forum, achieves traffic shaping, ensuring that traffic matches the service negotiated between users and the network during connection establishment.

Techniques to avoid congestion, recover from congestion, and ensure flow control in ATM available bit rate service include the use of resource management cells and switch behavior.


Previous Table of Contents Next

footer nav
Use of this site is subject certain Terms & Conditions.
Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details.