netdev - [RFC PATCH net-next 0/2] tcp: timer restart for tail loss

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-Id: <cover.1449478261.git.per.hurtig@kau.se>
Date:	Mon,  7 Dec 2015 10:00:10 +0100
From:	Per Hurtig <per.hurtig@....se>
To:	davem@...emloft.net, edumazet@...gle.com, ncardwell@...gle.com,
	nanditad@...gle.com, tom@...bertland.com, ycheng@...gle.com,
	viro@...iv.linux.org.uk, fw@...len.de, mleitner@...hat.com,
	daniel@...earbox.net, willemb@...gle.com,
	ilpo.jarvinen@...sinki.fi, pasi.sarolahti@....fi,
	stephen@...workplumber.org, netdev@...r.kernel.org
Cc:	anna.brunstrom@....se, apetlund@...ula.no, michawe@....uio.no,
	mohammad.rajiullah@....se, Per Hurtig <per.hurtig@....se>
Subject: [RFC PATCH net-next 0/2] tcp: timer restart for tail loss

This is a request for comments.

RTO and TLP restart is a modification to the restart process of the RTO and
TLP timers in TCP. Currently, both timers are restarted with its
corresponding timeout value when an acknowledgment (ACK) for correctly
received data is received. In many situations, resetting the timers on
incoming ACKs will add an implicit offset of at least RTT seconds to the
loss recovery process.

The goal of the modified restart is to provide quicker loss recovery for
segments lost in the end of a burst/connection, where the limited feedback
from a receiver inhibits the use of fast/early retransmit. To accomplish
this the algorithm adjusts the RTO and PTO (TLP's timer) values on each
rearm of the timers to allow them to expire after exactly RTO and PTO ms,
respectively.

The restart behavior is controlled by the new tcp_timer_restart sysctl.
tcp_timer_restart==0; disables RTOR and TLPR.
		 ==1; enables RTOR.
		 ==2; enables TLPR.
		 ==3; enables both RTOR and TLPR. [DEFAULT]

The new restart behavior has been approved by the IETF for publication as
an experimental RFC (for the RTO part of the mechanism) [1], and
experiments have been conducted to show both the benefit of this strategy
(in terms of reduced loss-recovery delays) and the limited negative impact
(in terms of spurious retransmissions) [2]. More information regarding the
mechanism can be found at [3].

Basic functionality tests, using packetdrill, are available at:
https://github.com/perhurt/packetdrill/tree/master/gtests/net/packetdrill/tests/linux/timer_restart

[1] https://datatracker.ietf.org/doc/draft-ietf-tcpm-rtorestart/
[2] http://www.sigcomm.org/sites/default/files/ccr/papers/2015/January/0000000-0000000.pdf
[3] http://riteproject.eu/resources/rto-restart/

Per Hurtig (2):
  tcp: RTO Restart (RTOR)
  tcp: TLP restart (TLPR)

 Documentation/networking/ip-sysctl.txt | 12 ++++++++++++
 include/net/tcp.h                      |  6 +++++-
 net/ipv4/sysctl_net_ipv4.c             | 10 ++++++++++
 net/ipv4/tcp_input.c                   | 26 +++++++++++++++++++++++++-
 net/ipv4/tcp_output.c                  | 12 ++++++++++--
 5 files changed, 62 insertions(+), 4 deletions(-)

-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html