[reti-accesso] carrier extension gigaethernet

[Giuseppe Bianchi < >]

At 17.57 23/02/2007, 
  wrote:
> Credo sia 50 us poiché il Sifs è
> 10 e lo slot è per motivi tecnologici 20us. E poiché 30 us servono per il
> PCF allora il primo tempo a disposizione maggiore di sifs  us è 50
> us.

Il notaio conferma: 50 (SIFS+2SLOT) perche' 30 (SIFS+1slot) e' usato per
il PFIS.


> Invece io non capisco la
> soluzione carrier extension di gigaethernet, chi la può
> spiegare?

Ecco un cut&paste dal libro - chiarissimo - dell'O-reilly dello
Spurgeon (che vi avevo consigliato ma non "imposto" - ecco
perche' non vi rimando al libro ma faccio io il cut&paste). Lo
trovate sul mulo, as usual.

Se dopo questa lettura le cose non fossero ancora chiare, ditemelo (ma
penso proprio di no!).

PS: l'autore ha un sito web su ethernet che e' ben fatto. Forse il link
era su qualche mia slide ma nel dubbio  eccolo di nuovo.

http://www.ethermanage.com/ethernet/ethernet.html









Gigabit Ethernet
Half-Duplex Network Diameter
A major challenge for
the engineers writing the Gigabit Ethernet standard was to provide a
sufficiently large network diameter in half-duplex mode. As we've seen,
the maximum network
diameter (i.e., cable distance) between any two stations largely
determines the slot time, which is an
essential part of the CSMA/CD MAC mechanism.
Repeaters, transceivers and the interfaces have circuits that require
some number of bit times to
operate. The combined set of these devices used on a network takes a
significant number of bit
times to handle frames, respond to collisions, and so on. It also takes a
small amount of time for a
signal to travel over a length of fiber optic or metallic cable. All of
this results in the total timing
budget for signal propagation through a system, which determines the
maximum cabling diameter
allowed when building a half-duplex Ethernet system.
Page 61
In Gigabit Ethernet, the
signaling happens ten times faster than it does in Fast Ethernet,
resulting in a
bit time that is one-tenth the size of the bit time in Fast Ethernet.
Without any changes in the timing
budget, the maximum network diameter of a Gigabit Ethernet system would
be about one-tenth of
that for Fast Ethernet, or in the neighborhood of 20 meters (65.6
feet).
Twenty meters is usable within a single room, such as a machine room
equipped with a set of
servers. However, one of the goals for the Gigabit Ethernet system is to
support a large enough
half-duplex network diameter to reach from a Gigabit Ethernet hub to the
desktop in a standard
office building. Desktop cabling for office buildings is typically based
on structured cabling
standards, which require the ability to reach 100 meters from a hub port.
This means that the total
network diameter may reach a maximum of 200 meters when connecting two
stations to a Gigabit
Ethernet repeater hub.*
Looking for Bit
Times
To meet the 200 meter
half-duplex network diameter goal, the designers of the Gigabit
Ethernet
system needed to increase the round-trip timing budget to accommodate
longer cables. If it was
somehow possible to speed up the internal operations of devices, such as
repeaters, then it might be
possible to increase the bit timing budget. The idea is that you could
save some bit times that are
consumed when signals are sent through repeater hubs. Unfortunately, with
today's technology, it is
not possible to produce repeaters and other components with ten times
less delay than equivalent
Fast Ethernet devices.
The next place you might think of looking to save bit times is in the
cable propagation delays.
However, it turns out that the signal propagation delay through the
cables cannot be reduced, as the
delay is fundamentally based on the speed of light (which is notoriously
difficult to improve).
Another way to gain time and achieve longer cable distance was in the
minimum frame transmission
specification. If the minimum frame time was extended then the Ethernet
signal would stay on the
channel longer. This would extend the round-trip time of the system and
make it possible to achieve
the 200 meter goal for the cabling diameter over twisted-pair cable. The
problem with this scheme,
however, is that changing the minimum frame length would make the frame
incompatible with the
other varieties of Ethernet—all of which use the same standard minimum
frame length.
* Repeaters can only be used in a half-duplex, shared Ethernet channel.
By definition, anything
connected to a repeater hub must be operating in half-duplex mode.
Page 62
Carrier
Extension
The solution to this
conundrum is to extend the amount of time occupied by the signal
carrier
associated with a minimum frame transmission, without actually modifying
the minimum frame length
or other frame fields. Gigabit Ethernet does this by extending the amount
of time a frame signal is
active on a half-duplex system with a mechanism called carrier extension,
as shown in Figure 3-2.
The frame signal, or Carrier, is extended by appending non-data
signals, called extension bits.
Extension bits are used when sending short frames so that the frame
signal stays on the system for a
minimum of 512 bytes (4,096 bit times), which is the new slot time for
Gigabit Ethernet. This slot
time makes it possible to use longer cables, and is also used in the
collision backoff calculations on
Gigabit Ethernet systems.
Figure 3-2.
Carrier extension
The use of carrier extension bits assumes that the underlying physical
signaling system is capable of
sending and receiving non-data symbols. Signaling for all fiber optic and
metallic cable Gigabit
Ethernet systems is based on signal encoding schemes that provide
non-data symbols that will trigger
carrier detection in all station transceivers. This makes it possible to
use these non-data symbols as
carrier extension bits without having the extension bits confused with
real frame data. These
encoding schemes are described in more detail in Chapter 6.
With carrier extension, a minimum size frame of 64 bytes (512 bits) is
sent on a Gigabit Ethernet
channel along with 448 extension bytes (3,584 bits), resulting in a
carrier signal that is 512 bytes in
length. Any frame less than 4,096 bits long will be extended as much as
necessary to provide carrier
for 4,096 bit times (but no more).
Carrier extension is a simple scheme for extending the collision domain
diameter; however, it adds
considerable overhead when transmitting short frames. A minimum-size
frame carrying 46 bytes of
data is 64 bytes in length. Carrier extension adds another 448 bytes of
non-data carrier extension
bits when the frame is transmitted, which significantly reduces the
channel efficiency when
transmitting short frames.
Page 63
The total impact on the
channel efficiency of a network will depend on the mix of frame sizes
seen
on the network. As the size of the frame being sent grows, the number of
extension bits needed
during transmission of the frame is reduced. When the frame being sent is
512 bytes or more in
length, no extension bits are used. Therefore, the amount of carrier
extension overhead encountered
when sending frames will vary depending on the frame size. Carrier
extension is only used in
half-duplex Gigabit Ethernet. In full-duplex mode the CSMA/CD MAC
protocol is not used, which
removes any concern about the slot time. Therefore, full-duplex Gigabit
Ethernet links do not need
carrier extension, and are able to operate at full efficiency.
Frame
Bursting
The Gigabit Ethernet
standard defines an optional capability called frame bursting to
improve
performance for short frames sent on half-duplex channels. This makes it
possible for a station to
send more than one frame during a given transmission event, improving the
efficiency of the system
for short frames. The total length of a frame burst is limited to 65,536
bit times plus the final frame
transmission, which sets a limit on the maximum burst transmission time.
Figure 3-3 shows how
frame bursting is organized.
Figure 3-3.
Frame bursting
Here's how the frame bursting system works. The first frame of the
transmission is always sent
normally, so that the first frame is sent by itself, and will be extended
if necessary. Because collision
always occurs within the first slot time, only this frame can be affected
by a collision, requiring it to
be retransmitted. This frame may encounter one or more collisions during
the transmission attempt.
Page 64
However, once the first frame
(including any extension bits) is transmitted without a collision, then
a
station equipped with frame bursting can keep sending additional frames
until the 65,536 bit time
burst limit is reached. To accomplish this, the transmitting station must
keep the channel from
becoming idle between frame transmissions. If the station became idle
during frame transmission,
other stations would try to acquire the channel, leading to
collisions.
The frame bursting station keeps the channel active by transmitting
special symbols that are
understood by all stations to be non-data symbols during the interframe
gap times of the frames. This
causes all other stations to continue to sense carrier (activity)
on the channel. This, in turn, causes
the other stations to continue deferring to passing traffic, allowing the
frame bursting station to
continue sending frames without concern that a collision will occur.
In essence, the first frame transmission clears the channel for the
subsequent burst frames. Once the
first frame has been successfully transmitted on a properly designed
network, the rest of the frames
in a burst are guaranteed not to encounter a collision. Frames sent
within a burst do not require
extension bits. The transmitting station is allowed to continue sending
frames in a burst until the
Frame Burst Limit (FBL) is reached, which is the time to the start
of the last frame in the burst.
For short frames, the optional frame bursting mechanism can improve the
utilization rate of the
channel. However, this can only occur if the station software is designed
to take advantage of the
ability to send bursts of frames. Without frame bursting, a half-duplex
Gigabit Ethernet channel is
less than twice as fast as a Fast Ethernet channel for the shortest frame
size. With frame bursting, the
Gigabit Ethernet channel is slightly over nine times faster than the Fast
Ethernet system for a constant
stream of short frames.
Frame bursting and channel efficiency
Without frame bursting the channel efficiency is low for a Gigabit
Ethernet channel carrying a
constant stream of 64-byte frames (512 bits). It requires an overhead of
one slot time to carry the
512 bit minimum size frame, which is 4096 bit times in the Gigabit
Ethernet system. To this we add
64 bit times of preamble and 96 bit times of interframe gap, for a total
of 4,256 bits of overhead.
Dividing the 512 bit payload by the overhead of 4,256 bits reveals a 12
percent channel efficiency.
With frame bursting the Gigabit Ethernet channel is considerably more
efficient for small frames,
since a whole series of frames can be sent in a burst without overhead
for the slot time once the
channel is acquired. Theoretically, you could send 93 small frames in a
single burst, with a channel
efficiency of over 90 percent. However, it is unlikely that the smallest
possible frames will dominate
the traffic
Page 65
flow in the real world. Nor is
it likely that a given station will have so many small frames to send
that
it can pack a frame burst full of them on a constant basis.
Remember, these limits only occur in half-duplex mode due to the
round-trip timing requirements.
Carrier extension is not needed in full-duplex mode, since full-duplex
mode does not use
CSMA/CD and is unconcerned about round-trip timing. A full-duplex Gigabit
Ethernet system can
operate at the full frame rate for all frame sizes, or ten times faster
than a full-duplex Fast Ethernet
system. Gigabit Ethernet performs quite well, particularly since the
vendors of Gigabit Ethernet
equipment support only the full-duplex mode of operation.