QoS
Cheat Sheet
IP QoS is *the* black
art of networking.
Before starting to
implement IP QoS you have to know in depth the theory of IP QoS, need
to know what can be achieve what cannot. It is not enough to know the
issues and the solutions in regards to IP QoS, because one thing is the
beautiful theory, another complete different thing is the practice.
No matter how good it
looks on paper, in practice we have to face a LOT of limitation of
networking devices (how network devices measure BW? Can do rate
limiting? Can do traffic shaping? Can guarantee min BW? Can guarantee
max BW? can be flexible? )
One thing for sure no
matter how good IP QoS works, it CANNOT guarantee that a packet will
arrive to the destination, because at the end of the day IP IS BEST
EFFORT protocol. Sorry to let you down, but it is because IP is best
effort make is so fantastically versatile ;)
In order to implement IP
QoS you have to know the theory of IP QoS in depth and then
Not only you need to
know in depth the theory of IP QoS before designing QoS
Benefits:
1. control over resource
-> preventing low-priority traffic from monopolizing link BW &
affecting high priority traffic
2. tailored service
-> provides offer of premium services along with best effort CoS
3. traffic
differentiation & provide service “guarantees”
4. more offecient use of
network resources
5. coexistence of
mission critical appls
IP is connectionless
end-to-end packet service protocol, provided best
effort:
- high
flexibility and robustness
-
prone to congestion
History:
- Type
of Service (ToS) byte in IP header but
ignored until lately
- The
Nagle Algorithm
- Van
Jacobson TCP slow start and congestion
avoidance (1986)
- TCP
fast retransmission and fast recovery for
optimal performance during period of packet loss (1990)
Level
of QoS:
1. Best-effort
service (no QoS):
- no guarantee as
to whether or when a packet is delivered to the
destination
- the only service the
Internet offers today
2. Differentiated
service (soft QoS):
- soft QoS <-- class based
- traffic is grouped info classes based on their service requirements
- statistical preferences not a hard guarantees
- allow a preferential treatment of one traffic over the other
- each traffic class is
differentiated by the network and serviced
according to the configured QoS mechanism for the class
- does not give service
guarantees, only differentiates traffic and
allow preferential treatment of one traffic class over the other
- works well for BW
intensive data applications
3. Guaranteed
service (hard QoS):
- hard QoS <--
required rigid guarantees from the network
- provide deterministic delay guarantees
- requires network resource reservation to
ensure the network meets a traffic flow’s specific service requirements.
- path reservation with a granularity of as single flow don’t scale
over Internet backbone <~ aggregated reservation <~ only a
minimum state of info in Internet core routers should be a scalable
means
Performane
Measures:
1. bandwidth: rate throughput
capacity of a given medium , protocol, or connection <-“size of
pipe”
2. packet delay: consist of
serialization delay, propagation delay, switching delay
3. jitter: variation in packet
delay
4. packet loss: number of
packets being lost by the network during transmission, generally occur
at congestion points
1. bandwidth: “size of
the pipe”
2. delay/latency:
2 parts: fixed & variable
a.
fixed-network delay: incl encoding & decoding time -> latency
required for electrical &/ optical signal to travel the media to
the receiver
~> QoS doesnst affect fixed-network delay
~> property of the medium, solution: upgrade
b. variable-network delay: refers to network condition -> congestion
~> applying QoS does affect
~> queueing delay
consist of.
>
packetizatn delay -+
>
serializatn delay |
>
propagatn delay +--> fixed
delay
>
processing delay |
>
switching delay -+
>
queueing delay ---> variable
delay
3. jitter: variable of
delay
~> buffer underrun
~> buffer overrun
4. packet loss: number
of packets being lost by the network during transmission
congestion --> packet drop.
Bit Error Rate (BER) of
10E-9 being relatively loss free
* classifying and
marking traffic --> network devices can differentiate traffic flow
* traffic conditioning
to tailor traffic flows to specific behavior and throughput
* marking traffic above
specific threshold as lower priority
* dropping packets when
rates reach specific threshold
* scheduling packet
--> higher-priority packets transmit from output queues before
lower-priority traffic
* managing output queues
--> avoid lower-priority packets waiting transmit monopolize buffer
space
* traffic management
--> token bucket <~ traffic metering scheme
* resource allocation
* congestion avoidance
& packet drop policy <~ active queue mgmt algorithm
enables routers to detect congestion before the queue overflows
* QoS signaling protocol
--> RSVP
* switching:
-
traditional: cached-based forwarding mech: efficient but scaling &
performance problem
- modern:
topology-based forwarding mech: building a forwarding table that
exactly matches the router’s routing table.
packet classifier and
marker:
- router at the
network’s edge user classifier to identify packet
belonging to a certain traffic class
marker:
- used
to color the classified traffic by setting
either IP Precedence or Differentiated Service Code Point (DSCP)
to
be
continued…
|