|
This page contains patches to Linux TCP implementation. The patches
have been used
in measurements performed in our research projects. The primary purpose
of the patches is to allow experimental use of TCP, thus it is
not intended to be used in
general networking. The code is very experimental in nature (read: not
beautiful or easily maintainable), and there may be bugs. Some of the
modifications, e.g. fixes on FRTO have been submitted to Linux network
development list and will be included to future releases, but many of
the modifications remain unsubmitted due to their nature. They can be
used for tests in closed network but not in a general Linux system.
|
patch-2.6.18.8-2007-05-21
|
Updated 2.6.18 series patch
|
|
patch-2.6.18.8-2007-04-19
|
This is the TCP patch for Linux 2.6.18 series.
The patch gives control over TCP implementation so
that different combination of RFCs may be enabled
at will. In addition, many of them allow just to
disabled the mainline Linux non-std behavior).
Complete list of tunable features:
- initial window,
- behavior when tcp is application limited
[RFC2861] (2.6.21 will apply a different
kind of fix to the same problem using
slow start after idle sysctl),
- delayed ACK timer auto-tuning,
- quick-ACK behavior (DAASS),
- ratehalving (Hoe's method),
- limited transmit [RFC3042],
- Control Block Interdependence [RFC2140]
(ssthresh, rtt,
reordering independently),
- F-RTO algorithm [RFC4138] (fixes + adds SACK
enhanced version; hits mainline at
2.6.22),
- retransmission timeout calculation [RFC2988]
(+min/max bounds),
- head segment timeouts (non-std
feature),
- use dup ACK to trigger recovery (Linux
uses SACK blocks by default),
- lost retransmission detection,
- dupThresh auto-tuning,
- SACK pipe calculation [RFC3517],
- advertized window auto-tuning,
- max-burst behavior, and
- RTO after RTO ssthresh halving
When using this patch, be sure to check that TCP
behaves as expected before jumping into conclusions.
|
|
patch-iip-2.6.14.4
|
This is the IIP patch for Linux kernel version
2.6.14.4. The patch gives control over
- initial window,
- behavior when tcp is application limited [RFC2861],
- delayed ack behavior,
- quick-ack behavior (DAASS),
- SACK pipe calculation [RFC3517],
- ratehalving (Hoe's method),
- limited transmit [RFC3042],
- Control Block Interdependence [RFC2140] (ssthresh, rtt,
reordering, complete disabling),
- retransmission timeout calculation [RFC2988], and
- F-RTO algorithm [RFC4138] (SACK enhanced version).
Limited transmit, RFC2988 RTT estimator and SACK enhanced F-RTO
have not been tested, so (more so than usual) when using
this patch, be sure to check that they behave as expected before
jumping into conclusions.
|
| patch-iip-12-2.4.19 |
This is the IIP patch for Linux kernel version 2.4.19. It works also with kernel 2.4.20. The patch gives control over the following functionalities:
- Control Block Interdependence (CBI) [RFC 2140]
- Forward RTO (F-RTO) [Internet draft: draft-sarolahti-tsvwg-tcp-frto-03.txt]
- Delayed acknowledgments
- Initial Window (Note: though the patch
does not limit this setting in any way, the value should not be set
larger than the recommended value)
- Limited transmit [RFC 3042]
- Rate halving [Internet draft: draft-ratehalving.txt]
Please read the readme-file for more information.
|
| patch-2.4.18-frto-20020419 |
This patch adds the F-RTO support to version
2.4.18 of Linux kernel. |
