Browse Prior Art Database

Method for minimum delay forwarding on network processors

IP.com Disclosure Number: IPCOM000109308D
Publication Date: 2005-Mar-23
Document File: 7 page(s) / 86K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method for minimum delay forwarding on network processors. Benefits include improved functionality and improved performance.

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 30% of the total text.

Method for minimum delay forwarding on network processors

Disclosed is a method for minimum delay forwarding on network processors. Benefits include improved functionality and improved performance.

Background

              Network processors conventionally use a store-and-forward architecture for forwarding packets. This mechanism facilitates processing between receive and transmit stages and is relatively easy to implement. However, the store-and-forward approach has a latency of more than 2 maximum transmission units (MTUs) in the worst case, where MTU is the maximum transmission unit. With the addition of protocols, a certain class of traffic typically transits through the system and undergoes minimum processing. This traffic may have a tight latency bound of 1 MTU or less associated with it.

              Conventional networking software has a minimum latency of 2 MTU per packet due to being based on store-and-forward mechanisms. A latency of 1 MTU is incurred to receive a packet that has a size of 1 MTU. A latency of 1MTU is incurred to transmit the packet. Therefore, the standard latency is 2 MTU. This analysis does not include the actual processing times within the network processor. The actual processing times are additional to the 2-MTU latency.

              Some classes of traffic undergo very minimum processing and may require a stringent latency bound of 1 MTU, as in the case of RPR Class A traffic.

General description

      The disclosed method provides a minimum delay of 1 MTU with a conventional store-and-forward mechanism. To overcome the latency barrier of 2 MTU and provide a 1-MTU latency for certain classes of traffic, a minimum-delay algorithm is added to the software architecture.

              The disclosed method includes a receive buffer (RBUF) element called an mpacket. Its size can be configured to 64, 128 or 256 bytes. Each mpacket is accompanied by a receive status word (RSW) from which the RX driver obtains the pointer to the mpacket residing in RBUF along with the mpacket’s start of packet (SOP), middle of packet (MOP), and end of packet (EOP) states. The RSW is automatically pushed to the transfer registers of the RX-driver thread at the start of the reception of an mpacket in an RBUF element.

      The key elements of the disclosed method include:

•     Receive (Rx) driver: the software on the network processor that receives the traffic

•     Transmit (Tx) driver: the software on the network processor that transmits the traffic

Advantages

      The disclosed method provides advantages, including:
•             Improved functionality due to providing a 1-MTU latency with a conventional store-and-forward packet processing mechanism

•     Improved functionality due to providing an RX packet-processing driver

•     Improved functionality due to providing a TX packet-processing driver

•     Improved functionality due to supporting store-and-forward traffic simultaneously with minim...