This is a handy tool which can help debugging any problems and should be
part of the distribution.
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
- Modification of iostat line as per input in Bug#12702
Fixes: 12702
Signed-off-by: Adolf Belka <adolf.belka@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
- Modification of iostat line as per input in Bug#12702
Fixes: 12702
Signed-off-by: Adolf Belka <adolf.belka@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
This change is necessary because we are using the right-hand two bytes
for storing the QoS classes.
All IPsec traffic will now be skipped and never classified by the QoS.
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
This will significantly reduce the load when classifying outgoing
traffic as there won't be any overhead as soon as the connection has
been classified. The classficiation is being stored in the iptables MARK
which will be copied to CONNMARK if changed.
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
This feature has to go in order to take advantage of CONNMARK which will
drastically decrease CPU load when passing packets.
We no longer will see every packet in the QOS-INC chain in order to
change classification of that packet. It is also party counter-intuitive
to have parts of one connection in one class and the corresponding ACK
packets in another.
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
In order to not deal with any marks from NAT and the IPS, this patch
adds masks to all places where packets are being marked for individual
QoS classes.
Instead of being able to use the "fw" match in tc, we have to use the
u32 to apply the mask.
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
In order to use the highest two bits for surciata bypass, we will need
to make sure that whenever we compare any other marks, we do not care
about anything else.
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
rockchip has a large bootloader so this also increase the gap between partitiontable
and fist partition to 16MB on aarch64
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
u-boot for nanopi r2s (rockchip rk3328) need dtc to build the image
so this adds dtc as build dependency for u-boot
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
Since the last update of sysstat, the output of iostat has changed and
the web user interface showed wrong values.
This is now being fixed in this patch.
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Acked-by: Adolf Belka <adolf.belka@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
This is not necessary on our systems and according to the documentation
will reduce code size of the allocator which will result in better
performance.
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Acked-by: Peter Müller <peter.mueller@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
This is a new type of metric to find out what resource is currently a
bottleneck for the whole system. We might use this for graphs.
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Acked-by: Peter Müller <peter.mueller@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
This feature is now disabled (was disabled on ARM before) as we do not
need it:
"Select this option to enable fine granularity task irq time accounting.
This is done by reading a timestamp on each transitions between softirq
and hardirq state, so there can be a small performance impact."
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Acked-by: Peter Müller <peter.mueller@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
This change is required to make the system respond faster to any
realtime events (sending or receiving data packets).
It will wake up at least one core 1000 times a second which will result
in finer timer granularity and make scheduling smoother. HTB for
example sends large packet bursts on each timer even to keep up data
rates which is not helpful for most applications.
The change might increase resource consumption and overhead slightly on
some systems, but since we are running in an idle-dyntick configuration,
we should not keep awake any cores that have not been awake before.
Signed-off-by: Michael Tremer <michael.tremer@ipfire.org>
Acked-by: Peter Müller <peter.mueller@ipfire.org>
Signed-off-by: Arne Fitzenreiter <arne_f@ipfire.org>
