- Distances:
- Local Area Networks covers an
office building
- Distances between any 2 nodes in a LAN ≤ 500 m
- Local Area Networks covers an
office building
- Consequence:
- The signal will take
≤ 500/200,000,000 =
2.5 × 10-6 sec
to reach
any other node in the network.
- When a node starts a transmission,
the start of the transmission
will reach all other nodes
in the LAN
before the whole message is transmitted
Example:
- Conclusion
- We can reduce the probability of collision by sensing the channel (transmission medium) before transmitting a message
- The signal will take
≤ 500/200,000,000 =
2.5 × 10-6 sec
to reach
any other node in the network.
- Slotted and unslotted:
- CSMA protocol can be
slotted or
unslotted
- Unslotted:
transmission
can begin any time (unrestraint)
- Slotted: Transmissions can only begin at the start of a slot
- CSMA protocol can be
slotted or
unslotted
- The general CSMA protocol:
- When a node has a message
to transmit,
it first
senses the broadcast channel
(or carrier)
to determine if it is
idle
(not used by some other node)
- If the carrier
is busy,
the node
will wait
and retry at a later time
If the carrier is idle, the node will attempt to transmit
How a node decides to wait and retry will be explain below...
- When a node has a message
to transmit,
it first
senses the broadcast channel
(or carrier)
to determine if it is
idle
(not used by some other node)
- Note:
- The name Carrier Sense Multiple Access
should be clear now...
- The protocol calls for
sensing the carrier (channel)
And it needs to try many times before an attempt will be successful (= multiple access)
- The name Carrier Sense Multiple Access
should be clear now...
- You may think that
"What's good for the goose is good for the gander"...
So why can't Aloha use channel sensing ?
- Purpose of sensing the channel
- The purpose of sensing the channel is to reduce the likelihood of collisions
Reason that CSMA use channel sensing
- Sensing the channel
will reduce
the likelihood of a collision
Because:
- The length (in time)
of a packet
is much longer
that the
propagation delay
from the sender to the sensor
So:
-
If the channel is idle,
the chance is very good
(in a LAN)
that nobody is presently transmitting.
If the channel is busy, the chance is very good (in a LAN) that somebody is presently transmitting.
- The length (in time)
of a packet
is much longer
that the
propagation delay
from the sender to the sensor
-
Examples that show carrier sensing in Aloha is useless
- Suppose a node senses the channel is
idle (no transmission)
Question:
-
Can this fact help
a node
reduce the likelihood
of a collision ?
NO !!!
A node can get a false channel idle indication:
Sequence of events:
- Node A starts a transmission.
The transmission takes time to reach node B
- Before A's transmission reaches B,
B
senses that the channel is idle
- Result:
-
transmissions from nodes A and B
collide at the main computer
even though B
senses that the
channel was clear
when it started the transmissions....
- Node A starts a transmission.
- Suppose a node senses the channel is
busy (there is a transmission)
Question:
-
Can this fact help
a node
reduce the likelihood
of a collision ?
I.e., will a collision certainly occur when a node transmits ?
NO !!!
A node can get a false channel busy indication:
Sequence of events:
- Node A starts transmitting a message.
Due to the large distance, A will have finished transmitting the message, before the wave front of the message arrives at B
- When the wave front of A's message arrives at
B, node B wants to transmit a message.
- Node B senses the channel and detect A's transmission.
So with carrier sensing, B concludes that it should wait.
- HOWEVER
-
If B would transmit right away, the message
of B would arrive at the main computer
without suffering from collision !!
That is because by the time that B's message arrives at the main computer, A's message has been cleared at the main computer (collision at the main computer occurs only when 2 transmission reach the main computer's antenna simultaneously)
- So B will get a
false indication
of an busy channel
When B transmits while the channel is busy, its transmission will NOT collide with A's at the main computer site !!!
- Node A starts transmitting a message.
- Suppose a node senses the channel is
idle (no transmission)
- Conclusion
- When you sense that the
channel is idle;
you may
cause a collision
if you transmit
- When you sense that the channel is busy; you may not cause a collision if you do transmit
In other words:
- you did not learn anything
(got any wiser)
by sensing the channel
- Put it bluntly:
- it's useless to sense the channel in Alaho...
- When you sense that the
channel is idle;
you may
cause a collision
if you transmit
- There are 2 (or 3)
types of CSMA protocols:
- Non-persistent CSMA
(the whimp)
- 1-persistent CSMA (the stalker)
(General case: p-persistent CSMA (the gambler))
- Non-persistent CSMA
(the whimp)
- The non-persistent CSMA:
go away when channel is
busy
- If the node
senses that
the carrier is busy,
it performs a
backoff procedure:
- The node selects a
random number X
- The node then
wait
X time units (slots)
and repeat the protocol from the start....
- The node selects a
random number X
- If the node senses
that the carrier is idle,
the node transmits
in the
next slot
After transmitting a message, the CSMA protocol terminates.
(Message do not need to be ACKed.)
- If the node
senses that
the carrier is busy,
it performs a
backoff procedure:
- Schematically:
- Design rationele of the non-persistent CSMA:
- The non-persistent CSMA
is a "pessimistic" protocol:
- "I senses the channel is busy...
Pessimist: I believe that the system is heavily loaded
I.e., there are other nodes out there that are also trying to transmit
- So I better come back later
(The random wait will minimize the likelihood that 2 or more nodes will start transmission at the same time...)
- "I senses the channel is busy...
- The non-persistent CSMA
is a "pessimistic" protocol:
- Strength of non-persistent CSMA
- Achieves good performance
in heavily loaded systems
- when there are many nodes that have messages to transmit (heavily loaded), they will "auto schedule" themselves with the random
- Achieves good performance
in heavily loaded systems
- Weakness of non-persistent CSMA
- Throughput
(# successful transmissions per time unit)
can be low
in a light load system
E.g.: if there is only one node waiting, it can safely transmit as soon as the current transmission terminates
- Throughput
(# successful transmissions per time unit)
can be low
in a light load system
- The 1-persistent CSMA protocol:
stalk with a
purpose....
- If the node
senses that the
carrier is
busy,
it wait (stalks)
until the channel becomes idle
- If the node
senses that the
carrier is
idle,
the node transmits
the message
in the next slot
After transmitting a message, the CSMA protocol terminates.
(Message do not need to be ACKed.)
- If the node
senses that the
carrier is
busy,
it wait (stalks)
until the channel becomes idle
- Schematically:
- NOTE:
- When 2 or more nodes are waiting for the current transmission to end (using 1-persistent CSMA), their transmissions will surely collide when the current transmission ends...
- Rationele of the 1-persistent CSMA:
- 1-persistent CSMA
is an "optimistic" protocol
- "I senses the channel is busy, and
I believe that the system is
lightly loaded
(I.e., I'm the only one out there that is waiting to transmit)....
So I'll just wait until this fellow is done and then take my turn"
- "I senses the channel is busy, and
I believe that the system is
lightly loaded
- 1-persistent CSMA
is an "optimistic" protocol
- Strength of the 1-persistent CSMA protocol:
- Performs very well in
lightly loaded systems
- I.e., when there are few nodes that have messages to transmit
A node that finds the channel busy will most likely be the only node waiting for the channel to become clear and it will start transmitting as soon as the current transmission is done and will not waste any time.
- Performs very well in
lightly loaded systems
- Strength of the 1-persistent CSMA protocol:
- Performance
will suck at heavy load
There will be an excessive number of collisions.
- Performance
will suck at heavy load
- The p-persistent CSMA protocol:
the gambler
(middle ground between the
non-persistent and
1-persistent CSMA)
- If the node
senses that the
carrier is
busy,
it wait (stalks)
until the channel becomes idle
- If the node
senses that the
carrier is
idle:
- The node flip a coin
(pick a random number)
- If coin comes up head
(random number < p),
transmits the message
in the next slot
Otherwise, try again from the start....
After transmitting a message, the CSMA protocol terminates.
(Message do not need to be ACKed.) - The node flip a coin
(pick a random number)
- If the node
senses that the
carrier is
busy,
it wait (stalks)
until the channel becomes idle
- Schematically:
- NOTE:
- When 2 or more nodes
are waiting for
the current
transmission to end
(using
p-persistent CSMA,
they will
only collide
when:
- 2 or more nodes pick a random number that is less than p
- When 2 or more nodes
are waiting for
the current
transmission to end
(using
p-persistent CSMA,
they will
only collide
when:
-
Rationele of the p-persistent CSMA:
- The p-persistent CSMA
is designed to
avoid the
collision problem
in 1-persistent CSMA
- By picking a random number, the system tries to ensure that at most 1 node will transmit
- The p-persistent CSMA
is designed to
avoid the
collision problem
in 1-persistent CSMA
- What is the "right value" for p ?
-
p must be set to a value
so that it is highly likely
that exactly one node
will pick a random number that is
less than p
For example, if on the average, there are 2 nodes that are waiting for a transmission to finish, then:
- set p = 1/2 = 0.5.
On the other hand, if on the average, there are 3 nodes that are waiting for a transmission to finish, then:
- set p = 1/3 = 0.3333.
And so on...
Conclusion:
- p must be dynamic
It's hard to estimate the number of nodes that is busy...
So the p-persistent CSMA protocol is not implemented as a real life protcol.
-
p must be set to a value
so that it is highly likely
that exactly one node
will pick a random number that is
less than p