Bug#864073: TCP offloads disabled by default

2017-06-03 Thread Michael Biebl
Hi Hans-Kristian

Am 04.06.2017 um 01:56 schrieb Hans-Kristian Bakke:

> Luckily the fix is already in 233 and it is only a simple one liner that
> I really think should be patched into Stretch to avoid people getting
> hard to diagnose distro specific performance issues when going from
> jessie to stretch or coming from other distros.

Thanks for making us aware of this issue and providing such a detailed
bug report. This looks indeed like something which should be fixed for
stretch.
I intend to pull the patch, I'm just not sure if it will still make it
into 9.0 given the impeding release or will have to wait for the first
point release 9.1


Regards,
Michael

-- 
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?



signature.asc
Description: OpenPGP digital signature


Bug#864073: TCP offloads disabled by default

2017-06-03 Thread Hans-Kristian Bakke
Package: systemd
Version: 232-23
​Severity: important​

With systemd 232 it got the ability to set the NIC offloads. These offloads
are not initialized correctly though, so the NIC ends up without fully
enabled TCP offloads.

Example from stretch defaults:
...
tcp-segmentation-offload: on
tx-tcp-segmentation: off < ???
tx-tcp-ecn-segmentation: off [fixed]
tx-tcp-mangleid-segmentation: off
tx-tcp6-segmentation: on
udp-fragmentation-offload: off [fixed]
generic-segmentation-offload: on
...

If I do the following (default in 231 and earlier):

  ethtool -K  tx-tcp-segmentation on

I get the correct version
...
tcp-segmentation-offload: on
tx-tcp-segmentation: on
tx-tcp-ecn-segmentation: off [fixed]
tx-tcp-mangleid-segmentation: off
tx-tcp6-segmentation: on
udp-fragmentation-offload: off [fixed]
generic-segmentation-offload: on
...

This causes major performance issues in high speed networks (1Gbit/s and
higher) or in weak embedded devices and it is going to be (by accident) the
default in Stretch!

Luckily the fix is already in 233 and it is only a simple one liner that I
really think should be patched into Stretch to avoid people getting hard to
diagnose distro specific performance issues when going from jessie to
stretch or coming from other distros.

Some links with info:
* https://github.com/systemd/systemd/issues/4650
* https://github.com/systemd/systemd/pull/4639
* https://sourceforge.net/p/e1000/mailman/message/35663858/

Regards,
Hans-Kristian Bakke