| re: Fast Ethernet
Fast Ethernet is a long twisting story, but I'll try to keep it simple!
I will also describe 100VG-AnyLAN, which started life calling itself
Fast Ethernet, but lost the marketing war for the name. As things look
right now, it is also losing the battle for the hearts and minds of
high speed network users everywhere........
Fast Ethernet refers to a 100 Mbps version of Ethernet, being
standardized in 802.3. The standard just went out to ballot, so it
is pretty close to being finished. You will see many product
announcements at Interop next week. The most visible companies in this
space are Grand Junction, 3Com, and Synoptics. Digital Semiconductor
recently announced a PCI to FAST Ethernet MAC chip, so we are also
seen as participating in this space.
Fast Ethernet scales the CSMA/CD protocol for operation at a MAC
data rate 100 Mbps. It has the following parts to it:
1) The MAC protocol is the same as 802.3 CSMA/CD - with the exception
that the Interpacket Gap has been reduced to 960 ns. The same packet
formats apply as were seen on the CSMA/CD network (64 to 1508 byte
packet lengths. little endian order in and out of the MAC).
2) There are multiple PHY/PMD combinations, designated as follows:
a) 100BASE-TX - this is a PHY/PMD for use with category 5 UTP and
150 ohm STP. It is essentially the same as the PMD developed for
FDDI on UTP, i.e. it is scrambled MLT-3 coding. It transmits on
one pair, and receives on another pair, with the crossover in the
repeater. The cable can be up to 100 meters long. RJ45 pinouts are
the same as for 10BASE-T.
b) 100BASE-T4 - This is a PHY/PMD for use with up to 100
meters of category 3 UTP cable. It can also be used on category
5 UTP. It uses all 4 pairs in the UTP cable. The data stream is
split into 3 parts, and each part is sent using a 3 level coding
scheme, using one pair of the cable for each stream. The reason it
needs 4 pairs is as follows. Two of pairs are bidirectional, i.e.,
either end of the point to point link can transmit on that pair. The
third and fourth pairs are similar to what happens in 10BASE-T,
i.e., one end transmits on one pair, the other end transmits on the
other pair. Collisions are detected when a transmitter sees energy
in the pair that it doesn't use, when it is transmitting (same
as is done by 10BASE-T).
c) 100BASE-FX - This is a 1300 nm based optical link taken from the
FDDI multimode PMD spec. It supports a 2 km distance, subject to a
timing restriction that we'll get into. It transmits on one fiber,
and receives on a second, with the crossover in the cabling (just
like FDDI).
3) Repeater spec - This spec is a little more complicated than you are
used to seeing in 10 Mbps enet. The spec defines two types of repeaters,
referred to as Class 1 and Class 2. Class 2 repeaters support only one
PMD type, and you can have no more than 2 of them in the network. In
addition, they can't be separated by any more than 5 meters of cable.
Class 1 repeaters can interconnect dissimilar PMD types (e.g. 100BASE-TX
and 100BASE-T4) and you can only have 1 of them in the network.
Since you don't get anything for free, you trade speed for distance
in a CSMA/CD network. So, for 10 times the data rate, you are limited
to a network of no longer than 205 meters between stations. In other
words, Fast Ethernet is a workgroup solution, which must be bridged
or routed to a backbone. You may have seen switch announcements from
Synoptics and Grand Junction talking about Full Duplex links. It is
possible to run 100BASE-TX and 100BASE-FX links in a full duplex mode,
when the repeaters are replaced by bridges/switches. This means that
all links are now point to point links, and, that 2 km fiber link now
becomes useful for connecting switches over a large area.
4) NWAY Signaling protocol - This is a protocol that is being
introduced into both 10BASE-T and 100BASE-TX and 100BASE-T4 stations.
The protocol works as follows:
a) Stations wakeup in 10BASE-T mode.
b) The NWAY protocol looks like a 10BASE-T linkbeat pulse to
older 10BASE-T stations, which then enable their links for
operation as 10BASE-T.
c) An NWAY capable station will then exchange information with
its neighbor, to negotiate whether they should operate in
one of 4 modes:
a) 10 Mbps, half duplex
b) 10 Mbps, full duplex
c) 100 Mbps, half duplex
d) 100 Mbps, full duplex
It is expected that adapters will be capable of both 10 and 100 Mbps
operation, similar to what was seen in the 4/16 Mb token ring market.
Note that full duplex 100 Mbps does not work for 100BASE-T4 stations,
since they need all four pairs to talk half duplex. 100BASE-T4 stations
could support full duplex 10 Mbps operation though, since they only
need 2 pairs for that functionality.
100VG-AnyLAN, aka Demand Priority, aka 100BASE-VG
100VG-AnyLAN was a competing solution for Fast Ethernet that was
brought to the IEEE by Hewlett Packard. Its formal name is 802.12
Demand Priority. It allowed Ethernet formatted frames to be scaled up
to a 100 Mbps data rate on category 3 UTP cabling. Along the way, IBM
joined in, and support for 802.5 frames was added to the proposal. After
a long nasty standards battle that you might have read about, IEEE 802
decided to create a new committee, 802.12, to standardize 100VG-AnyLAN.
Sounds similar to the 802.3/802.4/802.5 decision of years earlier.........
100VG-AnyLAN supports the Ethernet frame format, and also the 802.5
frame format, but NOT AT THE SAME TIME, i.e., the network is configured
for either 802.3 frames, or, 802.5 frames. Its initial goal was to
allow 100Mbps ethernet frames to run on category 3 UTP. It now has
PMDs for category 3 or 5 UTP using 4 pair, category 5 UTP and 150 ohm
UTP using 2 pair, 500 meters of fiber running at 850 nm, or, 2 km of
fiber running at 1300 nm.
100VG-AnyLAN is also an entirely new MAC protocol, which uses a
centrally located hub to grant access to the network. There are no
hard and fast rules in the standard for either the maximum station to
station distance, or, the maximum number of stations. There is a
performance impact as stations are added, as there is with any shared
media LAN, but simple configuration rules are not part of the standard.
Demand Priority is similar to token passing without the token. It
works as follows. A station signals a request to transmit
to the hub it is connected to. The hub scans its ports, looking
for transmission requests from the attached stations, and services them
in a round robin order. When it recognizes a request, the hub
grants the attached station the right to transmit ONE packet onto the
network. When that transmission is completed, the hub waits
for a period of time to allow new requests to be signaled, and then
grants the next round robin request. When hubs are chained together in
a hierarchy, the control of the transmission order is controlled by the
hub at the top of the tree. Transmission requests from the stations
bubble up to this central hub, which passes the right to transmit
back down the tree to the hub closest to the station. That hub services
all requests from its ports in round robin order, and then passes control
back to the central hub. In this manner, the right to transmit is passed
from station to station, based on who has a pending request.
When a station joins the network it undergoes a training period, during
which time the hub learns its address. When a packet is received, it
is buffered just long enough to decode the address, and send the
packet to the port on which the station is attached. This provides
some limited security.
Demand Priority gets its name from the way it that it supports two
priority levels, called normal and high priority. Normal priority
requests are not serviced until all high priority requests have been
serviced. If a normal priority request is pending for longer than 300
ms, the hub promotes the normal request to high priority, and it
is then serviced by the high priority round robin queue. In other
words, it demands priority.
The original intent of 100VG-AnyLAN was to support category 3 UTP. It
did this by splitting the 100 Mbps MAC data stream into 4 parts,
scrambling each part, coding each part with a a 5/6 code and then
tranmitting it using a 2 level code. So, there are 4 wires being used
at each time. The entire protocol is too elaborate to describe here, but
the key point is that all 4 wires are in use for a transmission. For
the 2 pair and fiber systems, the signaling protocol is designed slightly
differently.
Please send me mail with any information as to what customers
are asking for in this area. I am in the hub group, and we are watching
this very carefully. In some cases, FDDI will serve a customer's
needs if explained to them. In other cases, people are waiting for
ATM. In some cases, they may want Fast Ethernet. Any feeedback you have
as to customer questions/expectations around Fast Ethernet or 100VG-AnyLAN
would be much appreciated.
|
| >>there is aa marketing war between Fast Ethernet and 100VG-AnyLAN, but
>>we seems out of the market and focus on FDDI. when we compete with
>>Synoptic, my customer say that Synoptic's salesman tell them that FDDI
>>is going to end of life because slower than ATM and much expensive than
>>Fast Ethernet and 100VG-AnyLAN. is it true?
Well, my comments are worth as much as most analysts (less than $0.01)
but there are a couple of things in favor for FDDI:
1. It's a mature standard with tons of interoperable equipment.
2. FDDI-FDDI performance will beat 100Mb-100Mb Ethernet performance
any day of the week.
3. FDDI over UTP prices are extremely competitive with today's high
end Token-Ring adapters and should be competitive with Fast
Ethernet adapters.
4. While the 100Mbps Ethernet camps are slugging it out over which
one is better, customers have a huge choice today over FDDI
vendors, especially in the backbone space where FDDI clearly
beats out 100Mbps Ethernet.
5. While ATM deployment is expected to grow, there is still a LOT
of work that needs to be done to get native ATM services and
support in today's operating systems. Fast Ethernet, Fast
Token-Ring, etc have the advantage here in that shared LAN
technology is well understood and supported in today's NOS's,
protocol stacks, etc. ATM is going to be awesome once it's
native to the desktop, but it's going to take time to get there.
100Mbps Ethernet certainly has its place and if I had to take a guess,
I would say that 10/100 Fast Ethernet adapters are going to become very
popular, at least on the desktop. However, FDDI is not dead or dying...
in my opinion and before you accept comments otherwise, ask the hub vendor
if they support FDDI today. I'll bet you they say yes. :-)
>>And SynOptic say that ehternet switch technology provide full 10 Mb/s
>>bandwith to each LAN segment and it is faster than Digital's multiports
>>bridge.
This is utter garbage. The Synoptics, Kalpana, whatever, Ethernet
switches can't come close to matching the performance of the GIGAswitch.
If there's any question about the GIGAswitch, ask 3Com. They're trying
to figure out how to keep up with it with their brouters.
- Larry
|