TCP User Timeout Option UTO draftietftcpmtcputo04'txt - PowerPoint PPT Presentation

About This Presentation
Title:

TCP User Timeout Option UTO draftietftcpmtcputo04'txt

Description:

TCP User Timeout Option (UTO) draft-ietf-tcpm-tcp-uto-04.txt. Lars ... Peers exchange local, per-connection user timeouts through ... by Caitlin Bestler) ... – PowerPoint PPT presentation

Number of Views:44
Avg rating:3.0/5.0
Slides: 7
Provided by: fernan3
Learn more at: https://www.ietf.org
Category:

less

Transcript and Presenter's Notes

Title: TCP User Timeout Option UTO draftietftcpmtcputo04'txt


1
TCP User Timeout Option (UTO)draft-ietf-tcpm-tcp-
uto-04.txt
  • Lars Eggert (NEC)
  • Fernando Gont (UTN/FRH)
  • 67th IETF Meeting, San Diego, California, USA
  • November 5-10, 2006

2
Overview
  • Peers exchange local, per-connection user
    timeouts through advisory TCP option and adapt
    local user timeout accordingly
  • Motivation
  • longer UTO tolerate longer disconnections
  • shorter UTO less TCP state at busy servers
  • TCP mod, not policy for picking user timeouts
  • Adopted as WG item at the 61st IETF Meeting
    (Washington, DC, USA).

3
Main changes introduced in -03
  • Thorough review by Gorry Fairhurst
  • Rearranged the Introduction
  • It is now clear from the beginning what the
    document is about
  • Removed discussion of SO_RCVTIMEO and SO_SNDTIMEO
  • these parameters refer to reading/writing from/to
    socket buffers, rather than waiting for data to
    be acked, etc.
  • Clarified that the UTO option is disabled by
    default.
  • Could be enabled on a per-connection basis by a
    socket option, or on a system-wide basis by a
    toggle (e.g., sysctl)
  • Enforce lower limit of one RTO
  • Having a USER TIMEOUT of less than one RTO could
    be problematic

4
Main changes introduced in -04
  • Added advise that an UTO SHOULD be sent in the
    first segment sent after the SYN segment that
    initiated the 3WHS. (as suggested by Caitlin
    Bestler)
  • Particularly useful if the end that performed the
    passive OPEN does not record all the information
    included in the initial SYN (e.g., SYN cookies).
  • Clarified the impact on interoperability of not
    negotiating the option during the connection
    establishment phase (as suggested by Jamshid
    Mahdavi)
  • Potential of 3 of failures (i.e., it is okay in
    the vast majority of cases)
  • Those failures result from violating the
    requirement that TCP MUST ignore unknown options

5
Changes to be introduced in -05
  • Editorial tweaks suggested by Mark Allman
  • No major changes

6
Moving forward
  • We think the draft is ready for WGLC.
  • Any questions/comments?
Write a Comment
User Comments (0)
About PowerShow.com