ADDER AdderLink ALIF1000T User Manual page 31

Forwarding modes
In essence, the job of a layer 2 switch is to
transfer as fast as possible, data packets arriving
at one port out to another port as determined
by the destination address. This is known as data
forwarding and most switches offer a choice
of methods to achieve this. Choosing the most
appropriate forwarding method can often have a
sizeable impact on the overall speed of switching:
• Store and forward is the original method
and requires the switch to save each entire
data packet to buffer memory, run an error
check and then forward if no error is found (or
otherwise discard it).
• Cut-through was developed to address
the latency issues suffered by some store
and forward switches. The switch begins
interpreting each data packet as it arrives.
Once the initial addressing information has
been read, the switch immediately begins
forwarding the data packet while the remainder
is still arriving. Once all of the packet has been
received, an error check is performed and, if
necessary, the packet is tagged as being in
error. This checking 'on-the-fly' means that cut-
through switches cannot discard faulty packets
themselves. However, on receipt of the marked
packet, a host will carry out the discard process.
• Fragment-free is a hybrid of the above two
methods. It waits until the first 64 bits have
been received before beginning to forward
each data packet. This way the switch is more
likely to locate and discard faulty packets that
are fragmented due to collisions with other
data packets.
• Adaptive switches automatically choose
between the above methods. Usually they start
out as a cut-through switches and change to
store and forward or fragment-free methods
if large number of errors or collisions are
detected.
So which one to choose? The Cut-through method
has the least latency so is usually the best to use
with AdderLink Infinity units. However, if the
network components and/or cabling generate a lot
of errors, the Store and forward method should
probably be used. On higher end store and forward
switches, latency is rarely an issue.
Layer 2 and Layer 3: The OSI model
When discussing network switches, the terms
Layer 2 and Layer 3 are very often used. These
refer to parts of the Open System Interconnection
(OSI) model, a standardised way to categorise the
necessary functions of any standard network.
There are seven layers in the OSI model and these
define the steps needed to get the data created by
you (imagine that you are Layer 8) reliably down
LAYER 7 Application
LAYER 6 Presentation
LAYER 5 Session
LAYER 4 Transport
LAYER 3 Network
LAYER 2 Data Link
LAYER 1 Physical
Network connection
onto the transmission medium (the cable, optical
fibre, radio wave, etc.) that carries the data to
another user; to complete the picture, consider the
transmission medium is Layer 0. In general, think of
the functions carried out by the layers at the top as
being complex, becoming less complex as you go
lower down.
As your data travel down from you towards
the transmission medium (the cable), they are
successively encapsulated at each layer within a
new wrapper (along with a few instructions), ready
for transport. Once transmission has been made
to the intended destination, the reverse occurs:
Each wrapper is stripped away and the instructions
examined until finally only the original data are left.
So why are Layer 2 and Layer 3 of particular
importance when discussing AdderLink Infinity?
Because the successful transmission of data relies
upon fast and reliable passage through network
switches – and most of these operate at either
Layer 2 or Layer 3.
The job of any network switch is to receive each
incoming network packet, strip away only the first
few wrappers to discover the intended destination
then rewrap the packet and send it in the correct
direction.
In simplified terms, the wrapper that is added
at Layer 2 (by the sending system) includes the
physical address of the intended recipient system,
i.e. the unique MAC address (for example,
09:f8:33:d7:66:12) that is assigned to every
networking device at manufacture. Deciphering
recipients at this level is more straightforward
LAYER 7
than at Layer 3, where the address of the recipient
LAYER 6
is represented by a logical IP address (e.g.
LAYER 5
192.168.0.10) and requires greater knowledge
LAYER 4
of the surrounding network structure. Due to
their more complex circuitry, Layer 3 switches are
LAYER 3
more expensive than Layer 2 switches of a similar
LAYER 2
build quality and are used more sparingly within
LAYER 1
installations.
30