> Hi Guys,
> Trying to understand QoS, but got very confused. Understand the basics
> behind it but got do not understand how packets/frames are "tagged" for QoS
> Layer 3 QoS uses TOS bits in the IP Frame. Also DSCP extends the possible
> number of QoS "classes". Is this all OK?
> Where does 802.1p Layer 2 QoS come in as I see that some switches (Layer 2)
> i.e Catalyst 5000 have QoS enabled would this be 802.1p in the Ethernet
> In a real network what are advantages/disadvantages of Layer2/3 QoS.
Each layer defines its own QoS mechanisms unrelated to the other
layers. So 802.1p (now 802.1D) defines QoS mechanims at Layer 2
for IEEE 802 networks.
DiffServ defines the QoS mechanisms at Layer 3 for IP networks.
In the Layer 2 world, the QoS that a frame receives is defined
either by the priority bits in the VLAN tag, or based on the
physical port on which it entered the switch. The default
queueing mechanism in the switch is priority queueing, and a
switch may have up to 8 of these. However, the priority bits are
used to map frames to "traffic classes" and any queueing
behvaior is allowed as long as the packet ordering constraints
are met -- no reordering is allowed for a given (SA, DA, Priority).
There are no admission control mechanisms defined (such as policing
DiffServ on the other hand defines IP packet behavior in the router
to achieve QoS. Upon entering the router, a packet is classified
either based on its existing DSCP or based on other fields in the
packet. The packet then goes through policing, remarking (where
the DSCP may be updated based on the results of classification/policing),
and finally queueing.
How they tie together is implementation-specific. Basically,
in DiffServ, you have the ability to mark packets of a certain
class as having a higher drop preference. This would be done
at the router. When the packet enters an Ethernet switched
network interconnecting routers, the Ethernet switch typically
won't understand this and will treat all packets at the same
drop preference. The router is also responsible for mapping
the DSCP to an appropriate user priority in the 802.1Q header.
So basically, the mechanism at L3 is much more richer, but it
can only take advantage of the functionality allowed by L2 when
the packet is going between routers if the routers interconnected
using an L2 network (as opposed to being directly connected).
Enabling QoS on your switch will require some configuration
in the network of policies that determine how a frame is supposed
to receive a certain QoS. This will typically be done at the
You will only see an advantage to QoS if your network is somewhat
overloaded. In those instances, certain traffic can be prioritized
over others. This allows applications with time-critical traffic
(such as VoIP) to achieve acceptable quality even in the presence
of large amounts of non-time-critical traffic (such as file
transfers). If the network is always very lightly loaded, you
won't have much to gain from enabling QoS.
This is a very short overview. Other than implementing the basic
mechanisms, there's many different ways that vendors use to tie
the mechanisms at L2/L3 together. I'd suggest understanding the
basics by reading the relevant standards, and then understanding
what capabilities are provided by your vendor.