Quality of Service
From Wikipractice
Contents |
Introduction
In multimedia networks, it is very important to consider the application needs and manage their traffic according to them. In order to achieve this, we use a mechanism called Quality of Service (QoS). QoS is an end to end mechanism and has to be configured both on the LAN and the.
What is QoS
QoS is a set of tools used to manage bandwidth, minimize loss, delay and delay variation between devices in a network. QoS mechanisms works only on private networks: the Internet carry traffic on a best effort basis. QoS problematic depends on the type of technology used. Between sites, on a WAN, bandwidth availability is the most frequent concern. Within a campus infrastructure, on a LAN, buffer management issued dominates.
Trusted boundary
Most actual IP networks are using a Differentiated Service Code (DSCP) to mark the packets. The network uses this information to classify, mark, shape and police traffic, and to perform intelligent queuing. Traffic is classified and marked at the boundary of the network, based on different criteria like application port or IP addresses, or a combination of both of them. A network device has two choices when receiving a packet: he can trust the traffic from the remote equipment, assuming that the marking is correct, or he can reset the marking to default one if the remote device is not trusted. Inside the trusted boundary, all packets coming from all ports are trusted.
The gray zone in figure above represents the trusted QoS zone. Devices at its limit are responsible for marking packets with the right QoS.
Traffic prioritization
The prioritization mechanism allows to give more priority to some packets than to some others, according to rules defined for the network. As an example, voice traffic does not suffer delay and delay variation and must be delivered quickly: therefore this traffic will be classified in a high priority.
The network needs to know which packet belongs to which class of service in order to handle them properly. To do this, packets are marked, and the network will use these marks to put them into the correct classes, materialized by queues in telecommunication equipments.
Traffic characterization
When putting in place Quality of Service, it is important to define which traffic will be prioritized over which one. In order to do this, an analyses of the application to carry on the network has to be performed, and the result will list applications according to their priorities. However, only the application who really needs to be classified will be: if all application are classified as high priority, the result will be a best effort network.
As an example, applications can be classified as follow:
- IP Telephony: this application is very sensitive to packet loss and delay variation. It has to be put in the higher priority class (real time).
- Video conference: this application is less sensitive than IP telephony, but when too many packets are lost, it becomes unusable.
- ERP: these applications are becoming the central nervous system of the companies: therefore, their traffic needs to be prioritized.
- Internet access: this can be bandwidth consuming and can jeopardize other more important applications. It is advised to carry this traffic as best effort.
It is strongly advised to validate the application priorities with the management of the company. Changing the classification mechanisms after having installed the network can be a nightmare. This also avoid to class all application and services of the company as the highest one: doing so means having all application served as best effort.
Classe of service
Traffic classification
The classification of the different classes of traffic is done in filtering the different IP packets, and in tagging then with a number, called DSCP. Best practice suggest to proceed to packet marking at the entry of the network, on the devices located on the trusted boundary. The DSCP represents the priority of the packed and will be used by all network devices in the trusted boundary.
It is not advised to re-class packets on other places in the network as this will increase complexity of management: if traffic classification has to be changed, it will be done at the boundary, but the chances to update the special remarking rule implemented somewhere else will be forgotten, leading to annoying troubleshooting. It is, by the way, to avoid such problems that the three layer hierarchy has been developed.
Implementation of class of service
On the network devices, class of service are usually implemented in using queuing mechanisms. Outgoing packets are send in a queue according to their DSCP, and the dequeuing mechanism will make sure they will be sent onto the interface in respect of the QoS.
The queue also has some thresholds: when it get full, packets will be discarded.
Usually, the real time class, used for voice, is implemented via a strict priority mechanism: it means that as long as this queue is not empty, not packets from other queue will be set onto the interface. This is necessary due to voice which does not tolerate packet drop, jitter and delay. However, as voice traffic is low compared to data traffic, it works fine.
Limitation with real time class
As stated above, the real time class is implemented through a strict priority mechanism. According to the behavior of this, it is not advise to reserve more than 30 to 35% of the link bandwidth for real time traffic. If this limit is overpassed, the lower priority traffic will be impacted. In the extreme cases, performances could be strongly degraded event if the real time service works fine.
