NANOG 15 Agenda

For details about Multiprotocol Label Switching, please see the <A HREF="http://www.ietf.org/html.charters/mpls-charter.html" TARGET="_BLANK">Multiprotocol Label Switching (MPLS Charter)</A>

This BOF provides a forum for discussion of problems associated with managing hundreds of frequently changing access-lists and rate-limits, across multiple routers throughout a network, and by multiple policy managers.<BR> <BR> Background: There has been a lot of discussion on the NANOG list and the inet-access list about the increasing need for better traffic management through traffic shaping and filters. Whether the reason is to reduce the effects of smurf attacks, control the use of services to prevent customers from putting a server up without authorization, or manage the priority and amount of bandwidth given to customers, today\'s networks are rapidly outgrowing traditional bandwidth management methods.<BR> <BR> A static configuration of access-lists and rate-limits works when you are only making changes to a single router on an infrequent basis. These management techniques worked just fine when bandwidth could be managed in great swaths. With more customers having access to higher bandwidths through DSL, cable modems, and other technologies, network operators are having to build increasingly more complex bandwidth management policies that change more frequently and must be distributed throughout an entire network of routers.<BR> <BR> Topics to be discussed in this BOF include: discussion of the problem, managing hundreds of frequently-changing access-lists and rate-limits, distributing a traffic-management policy throughout a network, debugging/diagnosing problems with the policies, and distributing policy management amongst multiple managers.

This presentation examines a current implementation of RED and describes how it performs under conditions of varying congestion. We have gathered data in the form of link utilization, link discards, link latency, and average queue lengths under a variety of configurations. The examined links vary from T1 to T3 speeds, which illustrate customer egress and network core, respectively. The study provides evidence of network changes due to RED configuration. It also provides suggested configuration guidelines, as well as a brief discussion of WRED/congestion avoidance futures.