Browse Prior Art Database

Efficient Multicast Packet Buffering Scheme for Switch Devices

IP.com Disclosure Number: IPCOM000131949D
Publication Date: 2005-Nov-22
Document File: 3 page(s) / 58K

Publishing Venue

The IP.com Prior Art Database

Related People

Jeffrey A. Wilcox: AUTHOR [+2]

Abstract

Multicast schemes are intended to distribute the bandwidth requirements of data streams with multiple recipients by requiring a single transmission of each data packet on any given link in a network. This can have an adverse impact on switch design, as each switch must potentially replicate such data packet to each of its ports. This puts strains on buffer implementation, leading to increased demand for die-size, memory performance, as well as creating issues in terms of layout for such devices. The present invention applies the distributed multicast approach to the switch implementation. Ports of the switch are grouped by physical proximity, each group having a single buffer structure in which a single copy of each multicast packet is maintained. A secondary, shared buffer structure is provided to distribute incoming packets all ports to such groups as are required to complete all transmissions of these packets.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 25% of the total text.

Efficient Multicast Packet Buffering Scheme for Switch Devices

Jeffrey A. Wilcox, Ron L. Swartzentruber

Abstract

Multicast schemes are intended to distribute the bandwidth requirements of data streams with multiple recipients by requiring a single transmission of each data packet on any given link in a network.  This can have an adverse impact on switch design, as each switch must potentially replicate such data packet to each of its ports.  This puts strains on buffer implementation, leading to increased demand for die-size, memory performance, as well as creating issues in terms of layout for such devices.

The present invention applies the distributed multicast approach to the switch implementation.  Ports of the switch are grouped by physical proximity, each group having a single buffer structure in which a single copy of each multicast packet is maintained.  A secondary, shared buffer structure is provided to distribute incoming packets all ports to such groups as are required to complete all transmissions of these packets.

Problem

The problems introduced by multicast traffic are manifold.  If one simply replicates an incoming packet to each outbound port, memory requirements to provide the needed buffering rapidly become prohibitive.  Furthermore, internal bandwidth to provide such replication simply recreates the network problem which multicast is intended to solve.  On the other hand, as the ports provided by a single switch increase in number, the performance required of a single memory soon outstrip what memory can provide.  Alternatives of providing wider data paths may reduce the demands on memory performance, but only at the cost of routing congestion.  Furthermore, the lack of physical proximity amongst ports makes it necessary to add pipeline stages to the data path, increasing routing congestion and power consumption. 

To this must be added the variations in traffic flow amongst the ports.  Such variations require that the packet be buffered for greater lengths of time in order to satisfy all ports.  With a centralized buffering scheme, this must either lead to deeper buffers or to network congestion.  Increasing the buffer size lead to increases in die size and the cost associated with such increases, while only exacerbating the issues of physical distribution.

One further issue which is addressed herein is the potential need for multiple copies of the tables associated with multicast distribution.  These tables contain masks for each multicast routing tag that indicate which ports of the switch are to transmit a packet with that tag.  A single table could become a bottleneck when multicast traffic is received on many ports simultaneously.  Alternatively, the table must, like the central buffer itself, be designed for higher performance rates than is strictly necessary.  Without such high performance, multiple copies of the table must be maintained to avoid bottlenecks.  This once again leads to increases of...