Browse Prior Art Database

TCP GLOBAL SLOW START (SAME IP)

IP.com Disclosure Number: IPCOM000009781D
Original Publication Date: 2000-May-01
Included in the Prior Art Database: 2002-Sep-18
Document File: 2 page(s) / 89K

Publishing Venue

Motorola

Related People

Rod Averbuch: AUTHOR [+2]

Abstract

TCP has several mechanisms to operate efficiently over a Wide Area Network. One mechanism is slow start. The TCP based on RFC2001 builds its transmit volume (window size) in an exponential way (1,2,4,8..).

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 50% of the total text.

MOTOROLA

Technical Developments

TCP GLOBAL SLOW START (SAME IP)

by Rod Averbuch and Guy Romano

OVERVIEW

Patent Disclosure #39ISH: TCP Global Slow Start (same IP)

TCP has several mechanisms to operate efficiently over a Wide Area Network. One mechanism is slow start. The TCP based on RFC2001 builds its transmit volume (window size) in an exponential way (1,2,4,8..).

There are two reasons for this mechanism

1. In a congested network the slow start avoids additional congestion contribution by building the transmitted load gradually.

2. TCP utilizes an adaptive timer that builds its value based on previous round trip delay measurements. Starting slow enables the TCP to train its timer while the impact of wrong timing consideration is minimized by sending small amount of data.

3. Avoiding of packet loss in congested rowter.

PROBLEM

Traditional TCP based applications use one or few TCP sessions. For example FTP uses two ses-

HTTP TCP Midleware IP

sions. With the new HTTP protocol, every object transmission from a WEB site is involved in one TCP session. Therefore, one HTTP session could easily (with the new fancy WEB sites) end up with more than 10 TCP sessions. In this case the slow start overhead is very high. The overhead is an important factor especially in premium media such as wireless networks.

PRIOR ART

. RFC2001

. FTP software does not perform slow start.

They wait for the first TCP ACK and use the measurement for setting up initially timer value. After that the entire window is sent. This concept works well in lightly loaded systems with a constant round trip delay. However, in the iDEN example this scheme will not work properly. The delay variation dne to the dynamic PCH allocation will cause TCP time outs which will dramatically reduce the throughput.

. Another solution f...