Browse Prior Art Database

Receive Side Packet Reorganization and Reassembly to Facilitate Better Protocol Processing

IP.com Disclosure Number: IPCOM000009605D
Publication Date: 2002-Sep-04
Document File: 1 page(s) / 22K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method that reduces the number of packets delivered to an operating system for a given protocol, by combining them into a single packet for a given header. Benefits include a reduction in CPU utilization and packet protocol processing overhead.

This text was extracted from a Microsoft Word document.
This is the abbreviated version, containing approximately 72% of the total text.

Receive Side Packet Reorganization and Reassembly to Facilitate Better Protocol Processing

Disclosed is a method that reduces the number of packets delivered to an operating system for a given protocol, by combining them into a single packet for a given header. Benefits include a reduction in CPU utilization and packet protocol processing overhead.

Background

Although there is a mechanism to offload bulk TCP data on the sending side, there is currently no similar mechanism for the receiving side. On the sending side, with the TCP segment offload feature enabled, a wire speed can be achieved with 17% to 20% CPU utilization on a moderately powered machine. On the receiving side, the same wire speed uses close to 65% of the CPU for a single gigabit receive stream. Gigabit adapters are being pushed to desktops, but currently desktops are incapable of handling wire speeds at the gigabit rate.

General Description

The disclosed method changes TCP segment size, sequence numbers, and TCP ack processing in a way that is similar to large offloads for reassembly at the HW level. The SW protocol stack only sees a large packet being received, and this reduces per packet overhead during header processing. The SW stack touches data far less often than the CPU, and this improves cache locality for packet data processing. Additionally, it reduces ack from being transmitted to the remote station at the stack level, which also reduces CPU utilization.

For TCP/IP networks, ack is generated at th...