vyos/vyos-documentation
1
0
Fork 0
mirror of https://github.com/vyos/vyos-documentation.git synced 2025-10-26 08:41:46 +01:00
Code Issues Packages Projects Releases Wiki Activity
vyos-documentation/docs/qos.rst
Daniil Baturin e16cdd29a5
Merge pull request #277 from currite/qos-tune-01 
qos: tuning several little details
2020-07-02 07:15:49 +07:00

1185 lines
44 KiB
ReStructuredText
Raw Blame History

.. _qos:
***
QoS
***
The generic name of Quality of Service or Traffic Control involves
things like shaping traffic, scheduling or dropping packets, which
are the kind of things you may want to play with when you have, for
instance, a bandwidth bottleneck in a link and you want to somehow
prioritize some type of traffic over another.
tc_ is a powerful tool for Traffic Control found at the Linux kernel.
However, its configuration is often considered a cumbersome task.
Fortunately, VyOS eases the job through its CLI, while using ``tc`` as
backend.
How to make it work
===================
In order to have VyOS Traffic Control working you need to follow 2
steps:
1. **Create a traffic policy**.
2. **Apply the traffic policy to an interface ingress or egress**.
But before learning to configure your policy, we will warn you
about the different units you can use and also show you what *classes*
are and how they work, as some policies may require you to configure
them.
Units
=====
When configuring your traffic policy, you will have to set data rate
values, watch out the units you are managing, it is easy to get confused
with the different prefixes and suffixes you can use. VyOS will always
show you the different units you can use.
Prefixes
--------
They can be **decimal** prefixes.
.. code-block:: none
kbit (10^3) kilobit per second
mbit (10^6) megabit per second
gbit (10^9) gigabit per second
tbit (10^12) terabit per second
kbps (8*10^3) kilobyte per second
mbps (8*10^6) megabyte per second
gbps (8*10^9) gigabyte per second
tbps (8*10^12) terabyte per second
Or **binary** prefixes.
.. code-block:: none
kibit (2^10 = 1024) kibibit per second
mibit (2^20 = 1024^2) mebibit per second
gibit (2^30 = 1024^3) gibibit per second
tbit (2^40 = 1024^4) tebibit per second
kibps (1024*8) kibibyte (KiB) per second
mibps (1024^2*8) mebibyte (MiB) per second
gibps (1024^3*8) gibibyte (GiB) per second
tibps (1024^4*8) tebibyte (TiB) per second
Suffixes
--------
A *bit* is written as **bit**,
.. code-block:: none
kbit (kilobits per second)
mbit (megabits per second)
gbit (gigabits per second)
tbit (terabits per second)
while a *byte* is written as a single **b**.
.. code-block:: none
kbps (kilobytes per second)
mbps (megabytes per second)
gbps (gigabytes per second)
.. _classes:
Classes
=======
In the :ref:`creating_a_traffic_policy` section you will see that
some of the policies use *classes*. Those policies let you distribute
traffic into different classes according to different parameters you can
choose. So, a class is just a specific type of traffic you select.
The ultimate goal of classifying traffic is to give each class a
different treatment.
Matching traffic
----------------
In order to define which traffic goes into which class, you define
filters (that is, the matching criteria). Packets go through these matching rules
(as in the rules of a firewall) and, if a packet matches the filter, it
is assigned to that class.
In VyOS, a class is identified by a number you can choose when
configuring it.
.. note:: The meaning of the Class ID is not the same for every type of
policy. Normally policies just need a meaningless number to identify
a class (Class ID), but that does not apply to every policy.
The the number of a class in a Priority Queue it does not only
identify it, it also defines its priority.
.. code-block:: none
set traffic-policy <policy> <policy-name> class <class-ID> match <class-matching-rule-name>
In the command above, we set the type of policy we are going to
work with and the name we choose for it; a class (so that we can
differentiate some traffic) and an identifiable number for that class;
then we configure a matching rule (or filter) and a name for it.
A class can have multiple match filters:
.. code-block:: none
set traffic-policy shaper MY-SHAPER class 30 match HTTP
set traffic-policy shaper MY-SHAPER class 30 match HTTPs
A match filter can contain multiple criteria and will match traffic if
all those criteria are true.
For example:
.. code-block:: none
set traffic-policy shaper MY-SHAPER class 30 match HTTP ip protocol tcp
set traffic-policy shaper MY-SHAPER class 30 match HTTP ip source port 80
This will match TCP traffic with source port 80.
There are many parameters you will be able to use in order to match the
traffic you want for a class:
- **Ethernet (protocol, destination address or source address)**
- **Interface name**
- **IPv4 (DSCP value, maximum packet length, protocol, source address,**
**destination address, source port, destination port or TCP flags)**
- **IPv6 (DSCP value, maximum payload length, protocol, source address,**
**destination address, source port, destination port or TCP flags)**
- **Firewall mark**
- **VLAN ID**
When configuring your filter, you can use the ``Tab`` key to see the many
different parameters you can configure.
.. code-block:: none
vyos@vyos# set traffic-policy shaper MY-SHAPER class 30 match MY-FIRST-FILTER
Possible completions:
description Description for this match
> ether Ethernet header match
interface Interface name for this match
> ip Match IP protocol header
> ipv6 Match IPV6 header
mark Match on mark applied by firewall
vif Virtual Local Area Network (VLAN) ID for this match
As shown in the example above, one of the possibilities to match packets
is based on marks done by the firewall, `that can give you a great deal of flexibility`_.
You can also write a description for a filter:
.. code-block:: none
set traffic-policy shaper MY-SHAPER class 30 match MY-FIRST-FILTER description "My filter description"
.. note:: An IPv4 TCP filter will only match packets with an IPv4 header length of
20 bytes (which is the majority of IPv4 packets anyway).
.. note:: IPv6 TCP filters will only match IPv6 packets with no header extension, see
https://en.wikipedia.org/wiki/IPv6_packet#Extension_headers
Default
-------
Often you will also have to configure your *default* traffic in the same
way you do with a class. *Default* can be considered a class as it
behaves like that. It contains any traffic that did not match any
of the defined classes, so it is like an open class, a class without
matching filters.
Class treatment
---------------
Once a class has a filter configured, you will also have to define what
you want to do with the traffic of that class, what specific
Traffic-Control treatment you want to give it. You will have different
possibilities depending on the Traffic Policy you are configuring.
.. code-block:: none
vyos@vyos# set traffic-policy shaper MY-SHAPER class 30
Possible completions:
bandwidth Bandwidth used for this class
burst Burst size for this class (default: 15kb)
ceiling Bandwidth limit for this class
codel-quantum
fq-codel - Number of bytes used as 'deficit' (default 1514)
description Description for this traffic class
flows fq-codel - Number of flows (default 1024)
interval fq-codel - Interval (milliseconds) used to measure the delay (default 100)
+> match Class matching rule name
priority Priority for usage of excess bandwidth
queue-limit Maximum queue size (packets)
queue-type Queue type for this class
set-dscp Change the Differentiated Services (DiffServ) field in the IP header
target fq-codel - Acceptable minimum queue delay (milliseconds)
For instance, with :code:`set traffic-policy shaper MY-SHAPER class 30 set-dscp EF`
you would be modifying the DSCP field value of packets in that class to
Expedite Forwarding.
DSCP values as per :rfc:`2474` and :rfc:`4595`:
+---------+------------+--------+------------------------------+
| Binary | Configured | Drop | Description |
| value | value | rate | |
+=========+============+========+==============================+
| 101110 | 46 | - | Expedited forwarding (EF) |
+---------+------------+--------+------------------------------+
| 000000 | 0 | - | Best effort traffic, default |
+---------+------------+--------+------------------------------+
| 001010 | 10 | Low | Assured Forwarding(AF) 11 |
+---------+------------+--------+------------------------------+
| 001100 | 12 | Medium | Assured Forwarding(AF) 12 |
+---------+------------+--------+------------------------------+
| 001110 | 14 | High | Assured Forwarding(AF) 13 |
+---------+------------+--------+------------------------------+
| 010010 | 18 | Low | Assured Forwarding(AF) 21 |
+---------+------------+--------+------------------------------+
| 010100 | 20 | Medium | Assured Forwarding(AF) 22 |
+---------+------------+--------+------------------------------+
| 010110 | 22 | High | Assured Forwarding(AF) 23 |
+---------+------------+--------+------------------------------+
| 011010 | 26 | Low | Assured Forwarding(AF) 31 |
+---------+------------+--------+------------------------------+
| 011100 | 28 | Medium | Assured Forwarding(AF) 32 |
+---------+------------+--------+------------------------------+
| 011110 | 30 | High | Assured Forwarding(AF) 33 |
+---------+------------+--------+------------------------------+
| 100010 | 34 | Low | Assured Forwarding(AF) 41 |
+---------+------------+--------+------------------------------+
| 100100 | 36 | Medium | Assured Forwarding(AF) 42 |
+---------+------------+--------+------------------------------+
| 100110 | 38 | High | Assured Forwarding(AF) 43 |
+---------+------------+--------+------------------------------+
.. _embed:
Embedding one policy into another one
-------------------------------------
Often we need to embed one policy into another one. It is possible to do
so on classful policies, by attaching a new policy into a class. For
instance, you might want to apply different policies to the different
classes of a Round-Robin policy you have configured.
A common example is the case of some policies which, in order to be
effective, they need to be applied to an interface that is directly
connected where the bottleneck is. If your router is not
directly connected to the bottleneck, but some hop before it, you can
emulate the bottleneck by embedding your non-shaping policy into a
classful shaping one so that it takes effect.
You can configure a policy into a class through the ``queue-type``
setting.
.. code-block:: none
set traffic-policy shaper FQ-SHAPER bandwidth 4gbit
set traffic-policy shaper FQ-SHAPER default bandwidth 100%
set traffic-policy shaper FQ-SHAPER default queue-type fq-codel
As shown in the last command of the example above, the `queue-type`
setting allows these combinations. You will be able to use it
in many policies.
.. note:: Some policies already include other embedded policies inside.
That is the case of Shaper_: each of its classes use fair-queue
unless you change it.
.. _creating_a_traffic_policy:
Creating a traffic policy
=========================
VyOS lets you control traffic in many different ways, here we will cover
every possibility. You can configure as many policies as you want, but
you will only be able to apply one policy per interface and direction
(inbound or outbound).
Some policies can be combined, you will be able to embed_ a different
policy that will be applied to a class of the main policy.
.. hint:: If you are looking for a policy for your outbound traffic but
you do not know what policy you need, you might consider FQ-CoDel_ as
your multipurpose nearly-no-configuration low-delay fair-queue
policy; if delay does not worry you and you want to manually allocate
bandwidth shares to specific traffic, then you should consider
Shaper_.
Drop Tail
---------
| **Queueing discipline:** PFIFO (Packet First In First Out).
| **Applies to:** Outbound traffic.
This the simplest queue possible you can apply to your traffic. Traffic
must go through a finite queue before it is actually sent. You must
define how many packets that queue can contain.
When a packet is to be sent, it will have to go through that queue, so
the packet will be placed at the tail of it. When the packet completely
goes through it, it will be dequeued emptying its place in the queue and
being eventually handed to the NIC to be actually sent out.
Despite the Drop-Tail policy does not slow down packets, if many packets
are to be sent, they could get dropped when trying to get enqueued at
the tail. This can happen if the queue has still not been able to
release enough packets from its head.
This is the policy that requieres the lowest resources for the same
amount of traffic. But **very likely you do not need it as you cannot
get much from it. Sometimes it is used just to enable logging.**
.. cfgcmd:: set traffic-policy drop-tail <policy-name> queue-limit <number-of-packets>
Use this command to configure a drop-tail policy (PFIFO). Choose a
unique name for this policy and the size of the queue by setting the
number of packets it can contain (maximum 4294967295).
Fair Queue
----------
| **Queueing discipline:** SFQ (Stochastic Fairness Queuing).
| **Applies to:** Outbound traffic.
Fair Queue is a work-conserving scheduler which schedules the
transmission of packets based on flows, that is, it balances traffic
distributing it through different sub-queues in order to ensure
fairness so that each flow is able to send data in turn, preventing any
single one from drowning out the rest.
.. cfgcmd:: set traffic-policy fair-queue <policy-name>
Use this command to create a Fair-Queue policy and give it a name.
It is based on the Stochastic Fairness Queueing and can be applied to
outbound traffic.
In order to separate traffic, Fair Queue uses a classifier based on
source address, destination address and source port. The algorithm
enqueues packets to hash buckets based on those tree parameters.
Each of these buckets should represent a unique flow. Because multiple
flows may get hashed to the same bucket, the hashing algorithm is
perturbed at configurable intervals so that the unfairness lasts only
for a short while. Perturbation may however cause some inadvertent
packet reordering to occur. An advisable value could be 10 seconds.
One of the uses of Fair Queue might be the mitigation of Denial of
Service attacks.
.. cfgcmd:: set traffic-policy fair-queue <policy-name> hash-interval <seconds>`
Use this command to define a Fair-Queue policy, based on the
Stochastic Fairness Queueing, and set the number of seconds at which
a new queue algorithm perturbation will occur (maximum 4294967295).
When dequeuing, each hash-bucket with data is queried in a round robin
fashion. You can configure the length of the queue.
.. cfgcmd:: set traffic-policy fair-queue <policy-name> queue-limit <limit>
Use this command to define a Fair-Queue policy, based on the
Stochastic Fairness Queueing, and set the number of maximum packets
allowed to wait in the queue. Any other packet will be dropped.
.. note:: Fair Queue is a non-shaping (work-conserving) policy, so it
will only be useful if your outgoing interface is really full. If it
is not, VyOS will not own the queue and Fair Queue will have no
effect. If there is bandwidth available on the physical link, you can
embed_ Fair-Queue into a classful shaping policy to make sure it owns
the queue.
.. _FQ-CoDel:
FQ-CoDel
--------
| **Queueing discipline** Fair/Flow Queue CoDel.
| **Applies to:** Outbound Traffic.
The FQ-CoDel policy distributes the traffic into 1024 FIFO queues and
tries to provide good service between all of them. It also tries to keep
the length of all the queues short.
FQ-CoDel fights bufferbloat and reduces latency without the need of
complex configurations. It has become the new default Queueing
Discipline for the interfaces of some GNU/Linux distributions.
It uses a stochastic model to classify incoming packets into
different flows and is used to provide a fair share of the bandwidth to
all the flows using the queue. Each flow is managed by the CoDel
queuing discipline. Reordering within a flow is avoided since Codel
internally uses a FIFO queue.
FQ-CoDel is based on a modified Deficit Round Robin (DRR_) queue
scheduler with the CoDel Active Queue Management (AQM) algorithm
operating on each queue.
.. note:: FQ-Codel is a non-shaping (work-conserving) policy, so it
will only be useful if your outgoing interface is really full. If it
is not, VyOS will not own the queue and FQ-Codel will have no
effect. If there is bandwidth available on the physical link, you can
embed_ FQ-Codel into a classful shaping policy to make sure it owns
the queue. If you are not sure if you need to embed your FQ-CoDel
policy into a Shaper, do it.
FQ-CoDel is tuned to run ok with its default parameters at 10Gbit
speeds. It might work ok too at other speeds without configuring
anything, but here we will explain some cases when you might want to
tune its parameters.
When running it at 1Gbit and lower, you may want to reduce the
`queue-limit` to 1000 packets or less. In rates like 10Mbit, you may
want to set it to 600 packets.
If you are using FQ-CoDel embedded into Shaper_ and you have large rates
(100Mbit and above), you may consider increasing `quantum` to 8000 or
higher so that the scheduler saves CPU.
On low rates (below 40Mbit) you may want to tune `quantum` down to
something like 300 bytes.
At very low rates (below 3Mbit), besides tuning `quantum` (300 keeps
being ok) you may also want to increase `target` to something like 15ms
and increase `interval` to something around 150 ms.
.. cfgcmd:: set traffic-policy fq-codel <policy name> codel-quantum <bytes>
Use this command to configure an fq-codel policy, set its name and
the maximum number of bytes (default: 1514) to be dequeued from a
queue at once.
.. cfgcmd:: set traffic-policy fq-codel <policy name> flows <number-of-flows>
Use this command to configure an fq-codel policy, set its name and
the number of sub-queues (default: 1024) into which packets are
classified.
.. cfgcmd:: set traffic-policy fq-codel <policy name> interval <miliseconds>
Use this command to configure an fq-codel policy, set its name and
the time period used by the control loop of CoDel to detect when a
persistent queue is developing, ensuring that the measured minimum
delay does not become too stale (default: 100ms).
.. cfgcmd:: set traffic-policy fq-codel <policy-name> queue-limit <number-of-packets>`
Use this command to configure an fq-codel policy, set its name, and
define a hard limit on the real queue size. When this limit is
reached, new packets are dropped (default: 10240 packets).
.. cfgcmd:: set traffic-policy fq-codel <policy-name> target <miliseconds>`
Use this command to configure an fq-codel policy, set its name, and
define the acceptable minimum standing/persistent queue delay. This
minimum delay is identified by tracking the local minimum queue delay
that packets experience (default: 5ms).
Example
^^^^^^^
A simple example of an FQ-CoDel policy working inside a Shaper one.
.. code-block:: none
set traffic-policy shaper FQ-CODEL-SHAPER bandwidth 2gbit
set traffic-policy shaper FQ-CODEL-SHAPER 100%
set traffic-policy shaper FQ-CODEL-SHAPER fq-codel
Limiter
-------
| **Queueing discipline:** Ingress policer.
| **Applies to:** Inbound traffic.
Limiter is one of those policies that uses classes_ (Ingress qdisc is
actually a classless policy but filters do work in it).
The limiter performs basic ingress policing of traffic flows. Multiple
classes of traffic can be defined and traffic limits can be applied to
each class. Although the policer uses a token bucket mechanism
internally, it does not have the capability to delay a packet as a
shaping mechanism does. Traffic exceeding the defined bandwidth limits
is directly dropped. A maximum allowed burst can be configured too.
You can configure classes (up to 4090) with different settings and a
default policy which will be applied to any traffic not matching any of
the configured classes.
.. note:: In the case you want to apply some kind of **shaping** to your
**inbound** traffic, check the ingress-shaping_ section.
.. cfgcmd:: set traffic-policy limiter <policy-name> class <class ID> match <match-name> description <description>
Use this command to configure an Ingress Policer, defining its name,
a class identifier (1-4090), a class matching rule name and its
description.
Once the matching rules are set for a class, you can start configuring
how you want matching traffic to behave.
.. cfgcmd:: set traffic-policy limiter <policy-name> class <class-ID> bandwidth <rate>
Use this command to configure an Ingress Policer, defining its name,
a class identifier (1-4090) and the maximum allowed bandwidth for
this class.
.. cfgcmd:: set traffic-policy limiter <policy-name> class <class-ID> burst <burst-size>
Use this command to configure an Ingress Policer, defining its name,
a class identifier (1-4090) and the burst size in bytes for this
class (default: 15).
.. cfgcmd:: set traffic-policy limiter <policy-name> default bandwidth <rate>
Use this command to configure an Ingress Policer, defining its name
and the maximum allowed bandwidth for its default policy.
.. cfgcmd:: set traffic-policy limiter <policy-name> default burst <burst-size>
Use this command to configure an Ingress Policer, defining its name
and the burst size in bytes (default: 15) for its default policy.
.. cfgcmd:: set traffic-policy limiter <policy-name> class <class ID> priority <value>
Use this command to configure an Ingress Policer, defining its name,
a class identifier (1-4090), and the priority (0-20, default 20) in
which the rule is evaluated (the lower the number, the higher the
priority).
Network Emulator
----------------
| **Queueing discipline:** netem (Network Emulator) + TBF (Token Bucket Filter).
| **Applies to:** Outbound traffic.
VyOS Network Emulator policy emulates the conditions you can suffer in a
real network. You will be able to configure things like rate, burst,
delay, packet loss, packet corruption or packet reordering.
This could be helpful if you want to test how an application behaves
under certain network conditions.
.. cfgcmd:: set traffic-policy network-emulator <policy-name> bandwidth <rate>
Use this command to configure the maximum rate at which traffic will
be shaped in a Network Emulator policy. Define the name of the policy
and the rate.
.. cfgcmd:: set traffic-policy network-emulator <policy-name> burst <burst-size>
Use this command to configure the burst size of the traffic in a
Network Emulator policy. Define the name of the Network Emulator
policy and its traffic burst size (it will be configured through the
Token Bucket Filter qdisc). Default:15kb. It will only take effect if
you have configured its bandwidth too.
.. cfgcmd:: set traffic-policy network-emulator <policy-name> network-delay <delay>
Use this command to configure a Network Emulator policy defining its
name and the fixed amount of time you want to add to all packet going
out of the interface. The latency will be added through the
Token Bucket Filter qdisc. It will only take effect if you have
configured its bandwidth too. You can use secs, ms and us. Default:
50ms.
.. cfgcmd:: set traffic-policy network-emulator <policy-name> packet-corruption <percent>
Use this command to emulate noise in a Network Emulator policy. Set
the policy name and the percentage of corrupted packets you want. A
random error will be introduced in a random position for the chosen
percent of packets.
.. cfgcmd:: set traffic-policy network-emulator <policy-name> packet-loss <percent>`
Use this command to emulate packet-loss conditions in a Network
Emulator policy. Set the policy name and the percentage of loss
packets your traffic will suffer.
.. cfgcmd:: set traffic-policy network-emulator <policy-name> packet-reordering <percent>`
Use this command to emulate packet-reordering conditions in a Network
Emulator policy. Set the policy name and the percentage of reordered
packets your traffic will suffer.
.. cfgcmd:: set traffic-policy network-emulator <policy-name> queue-limit <limit>
Use this command to define the length of the queue of your Network
Emulator policy. Set the policy name and the maximum number of
packets (1-4294967295) the queue may hold queued at a time.
Priority Queue
--------------
| **Queueing discipline:** PRIO.
| **Applies to:** Outbound traffic.
The Priority Queue is a classful scheduling policy. It does not delay
packets (Priority Queue is not a shaping policy), it simply dequeues
packets according to their priority.
.. note:: Priority Queue, as other non-shaping policies, is only useful
if your outgoing interface is really full. If it is not, VyOS will
not own the queue and Priority Queue will have no effect. If there is
bandwidth available on the physical link, you can embed_ Priority
Queue into a classful shaping policy to make sure it owns the queue.
In that case packets can be prioritized based on DSCP.
Up to seven queues -defined as classes_ with different priorities- can
be configured. Packets are placed into queues based on associated match
criteria. Packets are transmitted from the queues in priority order. If
classes with a higher priority are being filled with packets
continuously, packets from lower priority classes will only be
transmitted after traffic volume from higher priority classes decreases.
.. note:: In Priority Queue we do not define clases with a meaningless
class ID number but with a class priority number (1-7). The lower the
number, the higher the priority.
As with other policies, you can define different type of matching rules
for your classes:
.. code-block:: none
vyos@vyos# set traffic-policy priority-queue MY-PRIO class 3 match MY-MATCH-RULE
Possible completions:
description Description for this match
> ether Ethernet header match
interface Interface name for this match
> ip Match IP protocol header
> ipv6 Match IPV6 header
mark Match on mark applied by firewall
vif Virtual Local Area Network (VLAN) ID for this match
As with other policies, you can embed_ other policies into the classes
(and default) of your Priority Queue policy through the ``queue-type``
setting:
.. code-block:: none
vyos@vyos# set traffic-policy priority-queue MY-PRIO class 3 queue-type
Possible completions:
fq-codel Fair Queue Codel
fair-queue Stochastic Fair Queue (SFQ)
drop-tail First-In-First-Out (FIFO)
priority Priority queueing based on DSCP
random-detect
Random Early Detection (RED)
.. cfgcmd:: set traffic-policy priority-queue <policy-name> class <class-ID> queue-limit <limit>`
Use this command to configure a Priority Queue policy, set its name,
set a class with a priority from 1 to 7 and define a hard limit on
the real queue size. When this limit is reached, new packets are
dropped.
.. _Random-Detect:
Random-Detect
-------------
| **Queueing discipline:** Generalized Random Early Drop.
| **Applies to:** Outbound traffic.
A simple Random Early Detection (RED) policy would start randomly
dropping packets from a queue before it reaches its queue limit thus
avoiding congestion. That is good for TCP connections as the gradual
dropping of packets acts as a signal for the sender to decrease its
transmission rate.
In contrast to simple RED, VyOS' Random-Detect uses a Generalized Random
Early Detect policy that provides different virtual queues based on the
IP Precedence value so that some virtual queues can drop more packets
than others.
This is achieved by using the first three bits of the ToS (Type of
Service) field to categorize data streams and, in accordance with the
defined precedence parameters, a decision is made.
IP precedence as defined in :rfc:`791`:
+------------+----------------------+
| Precedence | Priority |
+============+======================+
| 7 | Network Control |
+------------+----------------------+
| 6 | Internetwork Control |
+------------+----------------------+
| 5 | CRITIC/ECP |
+------------+----------------------+
| 4 | Flash Override |
+------------+----------------------+
| 3 | Flash |
+------------+----------------------+
| 2 | Immediate |
+------------+----------------------+
| 1 | Priority |
+------------+----------------------+
| 0 | Routine |
+------------+----------------------+
Random-Detect could be useful for heavy traffic. One use of this
algorithm might be to prevent a backbone overload. But only for TCP
(because dropped packets could be retransmitted), not for UDP.
.. cfgcmd:: set traffic-policy random-detect <policy-name> bandwidth <bandwidth>
Use this command to configure a Random-Detect policy, set its name
and set the available bandwidth for this policy. It is used for
calculating the average queue size after some idle time. It should be
set to the bandwidth of your interface. Random Detect is not a
shaping policy, this command will not shape.
.. cfgcmd:: set traffic-policy random-detect <policy-name> precedence <IP-precedence-value> average-packet <bytes>
Use this command to configure a Random-Detect policy and set its
name, then state the IP Precedence for the virtual queue you are
configuring and what the size of its average-packet should be
(in bytes, default: 1024).
.. note:: When configuring a Random-Detect policy: **the higher the
precedence number, the higher the priority**.
.. cfgcmd:: set traffic-policy random-detect <policy-name> precedence <IP-precedence-value> mark-probability <value>
Use this command to configure a Random-Detect policy and set its
name, then state the IP Precedence for the virtual queue you are
configuring and what its mark (drop) probability will be. Set the
probability by giving the N value of the fraction 1/N (default: 10).
.. cfgcmd:: set traffic-policy random-detect <policy-name> precedence <IP-precedence-value> maximum-threshold <packets>
Use this command to configure a Random-Detect policy and set its
name, then state the IP Precedence for the virtual queue you are
configuring and what its maximum threshold for random detection will
be (from 0 to 4096 packets, default: 18). At this size, the marking
(drop) probability is maximal.
.. cfgcmd:: set traffic-policy random-detect <policy-name> precedence <IP-precedence-value> minimum-threshold <packets>
Use this command to configure a Random-Detect policy and set its
name, then state the IP Precedence for the virtual queue you are
configuring and what its minimum threshold for random detection will
be (from 0 to 4096 packets). If this value is exceeded, packets
start being eligible for being dropped.
The default values for the minimum-threshold depend on IP precedence:
+------------+-----------------------+
| Precedence | default min-threshold |
+============+=======================+
| 7 | 16 |
+------------+-----------------------+
| 6 | 15 |
+------------+-----------------------+
| 5 | 14 |
+------------+-----------------------+
| 4 | 13 |
+------------+-----------------------+
| 3 | 12 |
+------------+-----------------------+
| 2 | 11 |
+------------+-----------------------+
| 1 | 10 |
+------------+-----------------------+
| 0 | 9 |
+------------+-----------------------+
.. cfgcmd:: set traffic-policy random-detect <policy-name> precedence <IP-precedence-value> queue-limit <packets>
Use this command to configure a Random-Detect policy and set its
name, then name the IP Precedence for the virtual queue you are
configuring and what the maximum size of its queue will be (from 1 to
1-4294967295 packets). Packets are dropped when the current queue
length reaches this value.
If the average queue size is lower than the **min-threshold**, an
arriving packet will be placed in the queue.
In the case the average queue size is between **min-threshold** and
**max-threshold**, then an arriving packet would be either dropped or
placed in the queue, it will depend on the defined **mark-probability**.
If the current queue size is larger than **queue-limit**,
then packets will be dropped. The average queue size depends on its
former average size and its current one.
If **max-threshold** is set but **min-threshold is not, then
**min-threshold** is scaled to 50% of **max-threshold**.
In principle, values must be
:code:`min-threshold` < :code:`max-threshold` < :code:`queue-limit`.
Rate Control
------------
| **Queueing discipline:** Tocken Bucket Filter.
| **Applies to:** Outbound traffic.
Rate-Control is a classless policy that limits the packet flow to a set
rate. It is a pure shaper, it does not schedule traffic. Traffic is
filtered based on the expenditure of tokens. Tokens roughly correspond
to bytes.
Short bursts can be allowed to exceed the limit. On creation, the
Rate-Control traffic is stocked with tokens which correspond to the
amount of traffic that can be burst in one go. Tokens arrive at a steady
rate, until the bucket is full.
.. cfgcmd:: set traffic-policy rate-control <policy-name> bandwidth <rate>
Use this command to configure a Rate-Control policy, set its name
and the rate limit you want to have.
.. cfgcmd:: set traffic-policy rate-control <policy-name> burst <burst-size>
Use this command to configure a Rate-Control policy, set its name
and the size of the bucket in bytes which will be available for
burst.
As a reference: for 10mbit/s on Intel, you might need at least 10kbyte
buffer if you want to reach your configured rate.
A very small buffer will soon start dropping packets.
.. cfgcmd:: set traffic-policy rate-control <policy-name> latency
Use this command to configure a Rate-Control policy, set its name
and the maximum amount of time a packet can be queued (default: 50
ms).
Rate-Control is a CPU-friendly policy. You might consider using it when
you just simply want to slow traffic down.
.. _DRR:
Round Robin
-----------
| **Queueing discipline:** Deficit Round Robin.
| **Applies to:** Outbound traffic.
The round-robin policy is a classful scheduler that divides traffic in
different classes_ you can configure (up to 4096). You can embed_ a
new policy into each of those classes (default included).
Each class is assigned a deficit counter (the number of bytes that a
flow is allowed to transmit when it is its turn) initialized to quantum.
Quantum is a parameter you configure which acts like a credit of fix
bytes the counter receives on each round. Then the Round-Robin policy
starts moving its Round Robin pointer through the queues. If the deficit
counter is greater than the packet's size at the head of the queue, this
packet will be sent and the value of the counter will be decremented by
the packet size. Then, the size of the next packet will be compared to
the counter value again, repeating the process. Once the queue is empty
or the value of the counter is insufficient, the Round-Robin pointer
will move to the next queue. If the queue is empty, the value of the
deficit counter is reset to 0.
At every round, the deficit counter adds the quantum so that even large
packets will have their opportunity to be dequeued.
.. cfgcmd:: set traffic-policy round-robin <policy name> class
<class-ID> quantum <packets>
Use this command to configure a Round-Robin policy, set its name, set
a class ID, and the quantum for that class. The deficit counter will
add that value each round.
.. cfgcmd:: set traffic-policy round-robin <policy name> class
<class ID> queue-limit <packets>
Use this command to configure a Round-Robin policy, set its name, set
a class ID, and the queue size in packets.
As with other policies, Round-Robin can embed_ another policy into a
class through the ``queue-type`` setting.
.. code-block:: none
vyos@vyos# set traffic-policy round-robin DRR class 10 queue-type
Possible completions:
fq-codel Fair Queue Codel
fair-queue Stochastic Fair Queue (SFQ)
drop-tail First-In-First-Out (FIFO)
priority Priority queueing based on DSCP
.. _Shaper:
Shaper
------
| **Queueing discipline:** Hierarchical Token Bucket.
| **Applies to:** Outbound traffic.
The Shaper policy does not guarantee a low delay, but it does guarantee
bandwidth to different traffic classes and also lets you decide how to
allocate more traffic once the guarantees are met.
Each class can have a guaranteed part of the total bandwidth defined for
the whole policy, so all those shares together should not be higher
than the policy's whole bandwidth.
If guaranteed traffic for a class is met and there is room for more
traffic, the ceiling parameter can be used to set how much more
bandwidth could be used. If guaranteed traffic is met and there are
several classes willing to use their ceilings, the priority parameter
will establish the order in which that additional traffic will be
allocated. Priority can be any number from 0 to 7. The lower the number,
the higher the priority.
.. cfgcmd:: set traffic-policy shaper <policy-name> bandwidth <rate>
Use this command to configure a Shaper policy, set its name
and the maximum bandwidth for all combined traffic.
.. cfgcmd:: set traffic-policy shaper <policy-name> class <class-ID> bandwidth <rate>
Use this command to configure a Shaper policy, set its name, define
a class and set the guaranteed traffic you want to allocate to that
class.
.. cfgcmd:: set traffic-policy shaper <policy-name> class <class-ID> burst <bytes>
Use this command to configure a Shaper policy, set its name, define
a class and set the size of the `tocken bucket`_ in bytes, which will
be available to be sent at maximum speed (default: 15Kb).
.. cfgcmd:: set traffic-policy shaper <policy-name> class <class-ID> ceiling <bandwidth>
Use this command to configure a Shaper policy, set its name, define
a class and set the maximum speed possible for this class. The
default ceiling value is the bandwidth value.
.. cfgcmd:: set traffic-policy shaper <policy-name> class <class-ID> priority <0-7>
Use this command to configure a Shaper policy, set its name, define
a class and set the priority for usage of available bandwidth once
guarantees have been met. The lower the priority number, the higher
the priority. The default priority value is 0, the highest priority.
As with other policies, Shaper can embed_ other policies into its
classes through the ``queue-type`` setting and then configure their
parameters.
.. code-block:: none
vyos@vyos# set traffic-policy shaper HTB class 10 queue-type
Possible completions:
fq-codel Fair Queue Codel
fair-queue Stochastic Fair Queue (SFQ)
drop-tail First-In-First-Out (FIFO)
priority Priority queueing based on DSCP
random-detect
Random Early Detection (RED)
.. code-block:: none
vyos@vyos# set traffic-policy shaper HTB class 10
Possible completions:
bandwidth Bandwidth used for this class
burst Burst size for this class (default: 15kb)
ceiling Bandwidth limit for this class
codel-quantum
fq-codel - Number of bytes used as 'deficit' (default 1514)
description Description for this traffic class
flows fq-codel - Number of flows (default 1024)
interval fq-codel - Interval (milliseconds) used to measure the delay (default 100)
+> match Class matching rule name
priority Priority for usage of excess bandwidth
queue-limit Maximum queue size (packets)
queue-type Queue type for this class
set-dscp Change the Differentiated Services (DiffServ) field in the IP header
target fq-codel - Acceptable minimum queue delay (milliseconds)
Example
^^^^^^^
A simple example of Shaper using priorities.
.. code-block:: none
set traffic-policy shaper MY-HTB bandwidth '50mbit'
set traffic-policy shaper MY-HTB class 10 bandwidth '10%'
set traffic-policy shaper MY-HTB class 10 ceiling '15%'
set traffic-policy shaper MY-HTB class 10 match ADDRESS10 ip source address '192.168.10.0/24'
set traffic-policy shaper MY-HTB class 10 priority '0'
set traffic-policy shaper MY-HTB class 10 queue-type 'fair-queue'
set traffic-policy shaper MY-HTB class 20 bandwidth '10%'
set traffic-policy shaper MY-HTB class 20 ceiling '50%'
set traffic-policy shaper MY-HTB class 20 match ADDRESS20 ip source address '192.168.20.0/24'
set traffic-policy shaper MY-HTB class 20 priority '3'
set traffic-policy shaper MY-HTB class 20 queue-type 'fair-queue'
set traffic-policy shaper MY-HTB class 30 bandwidth '10%'
set traffic-policy shaper MY-HTB class 30 ceiling '50%'
set traffic-policy shaper MY-HTB class 30 match ADDRESS30 ip source address '192.168.30.0/24'
set traffic-policy shaper MY-HTB class 30 priority '5'
set traffic-policy shaper MY-HTB class 30 queue-type 'fair-queue'
set traffic-policy shaper MY-HTB default bandwidth '10%'
set traffic-policy shaper MY-HTB default ceiling '100%'
set traffic-policy shaper MY-HTB default priority '7'
set traffic-policy shaper MY-HTB default queue-type 'fair-queue'
.. _ingress-shaping:
The case of ingress shaping
===========================
| **Applies to:** Inbound traffic.
For the ingress traffic of an interface, there is only one policy you
can directly apply, a **Limiter** policy. You cannot apply a shaping
policy directly to the ingress traffic of any interface because shaping
only works for outbound traffic.
This workaround lets you apply a shaping policy to the ingress traffic
by first redirecting it to an in-between virtual interface
(`Intermediate Functional Block`_). There, in that virtual interface,
you will be able to apply any of the policies that work for outbound
traffic, for instance, a shaping one.
That is how it is possible to do the so-called "ingress shaping".
.. code-block:: none
set traffic-policy shaper MY-INGRESS-SHAPING bandwidth 1000kbit
set traffic-policy shaper MY-INGRESS-SHAPING default bandwidth 1000kbit
set traffic-policy shaper MY-INGRESS-SHAPING default queue-type fair-queue
set interfaces input ifb0 traffic-policy out MY-INGRESS-SHAPING
set interfaces ethernet eth0 redirect ifb0
Applying a traffic policy
=========================
Once a traffic-policy is created, you can apply it to an interface:
.. code-block:: none
set interfaces etherhet eth0 traffic-policy out WAN-OUT
You can only apply one policy per interface and direction, but you can
have several policies working at the same time:
.. code-block:: none
set interfaces ethernet eth0 traffic-policy in WAN-IN
set interfaces etherhet eth0 traffic-policy out WAN-OUT
set interfaces etherhet eth1 traffic-policy out WAN-OUT
set interfaces ethernet eth2 traffic-policy out LAN-IN
set interfaces ethernet eth2 traffic-policy out LAN-OUT
Getting queueing information
----------------------------
.. opcmd:: show queueing <interface-type> <interface-name>
Use this command to see the queueing information for an interface.
You will be able to see a packet counter (Sent, Dropped, Overlimit
and Backlog) per policy and class configured.
.. _that can give you a great deal of flexibility: https://blog.vyos.io/using-the-policy-route-and-packet-marking-for-custom-qos-matches
.. _tc: https://en.wikipedia.org/wiki/Tc_(Linux)
.. _tocken bucket: https://en.wikipedia.org/wiki/Token_bucket
.. _HFSC: https://en.wikipedia.org/wiki/Hierarchical_fair-service_curve
.. _Intermediate Functional Block: https://www.linuxfoundation.org/collaborate/workgroups/networking/ifb
Reference in New Issue View Git Blame Copy Permalink