Very Computer

by **WONG SAI-KE** » Fri, 06 Apr 2001 09:31:05

I asked this question because I'm studying the networking in details.

Over 10 years ago when I read books about networking, mostly they
cannot provide solid idea about networking. Until recent years, I
found some books describing bits and pieces of what is actually using.
Even you will not be able to successfully config your network by
reading one book, after you read several book plus some hands on,
you'll get it.

Still, if you want to know what is doing inside a repeater, bridge, ...
there are rare books telling you about this. They always refer such
devices to OSI 7 layer model and full stop. But its hardly for you
to know what would be the behaviour of the devices when you get to
work with them.

My question is what is a repeater for 10base2 ? Because the repeater
is connected to 2 segments of co-axial, if there is signal from 1
segment, I can imagine the repeater will copy it to the other side.
but what if the other side also started to send ? If it copied,
thats OK coz eventualy both side should detect collision. Then it
must distinguish if the signal is send by itself or from the LAN
segment, that means the repeater should have a bit detection
rather than simply 2 amplifiers.

Further, the 10base2 rely on the characteristic impedance of the
50 ohm co-axial cable plus the maximum distance of 185M plus the
minimum 46 data bytes which sum up to 576 bits of stream to detect
the collision. (Anyone know how to interpret this calculation of
propagation delay ?) The specification said up to maximum of 5
segments with 4 repeater is possible. Is this collision scheme
base the calculation of propagation delay on full length for 185
X 5 plus repeater delay ?

And how is the hub for 10baseT operate ? Was the 10baseT designed
for passive hub ? Or only active hub is possible ? For passive
hub, I mean only passive electronics components are used, no
electric power is necessary.

Why when you cascade the hubs, you need an MDI/MDIX switch for a
specific port ? How many hubs can be chains at maximum ?

Thanks a lot for your patient to read the long question.

SK

by **James Knot** » Sat, 07 Apr 2001 22:59:26

> I asked this question because I'm studying the networking in details.

> Over 10 years ago when I read books about networking, mostly they
> cannot provide solid idea about networking. Until recent years, I
> found some books describing bits and pieces of what is actually using.
> Even you will not be able to successfully config your network by
> reading one book, after you read several book plus some hands on,
> you'll get it.

> Still, if you want to know what is doing inside a repeater, bridge, ...
> there are rare books telling you about this. They always refer such
> devices to OSI 7 layer model and full stop. But its hardly for you
> to know what would be the behaviour of the devices when you get to
> work with them.

> My question is what is a repeater for 10base2 ? Because the repeater
> is connected to 2 segments of co-axial, if there is signal from 1
> segment, I can imagine the repeater will copy it to the other side.
> but what if the other side also started to send ? If it copied,
> thats OK coz eventualy both side should detect collision. Then it
> must distinguish if the signal is send by itself or from the LAN
> segment, that means the repeater should have a bit detection
> rather than simply 2 amplifiers.

Repeaters regenerate the signal, rather than just amplify it. Also,
while they work in both directions, they do so in only one direction at
a time.

Quote:
> Further, the 10base2 rely on the characteristic impedance of the
> 50 ohm co-axial cable plus the maximum distance of 185M plus the
> minimum 46 data bytes which sum up to 576 bits of stream to detect
> the collision. (Anyone know how to interpret this calculation of
> propagation delay ?) The specification said up to maximum of 5
> segments with 4 repeater is possible. Is this collision scheme
> base the calculation of propagation delay on full length for 185
> X 5 plus repeater delay ?

The segment length is determined by signal loss etc. The overall
length, using multiple segments is determined by the minimum packet time
i.e. smallest packet, travelling to the furthest point of the network.
The minimum time is determined by the number of bits in the smallest
allowable packet * 0.1 uS per bit. The maximum distance is determined
by how far the signal will travel through the cable in that time. The
speed of the signal = the speed of light * the velocity factor of the
cable. The velocity factor is alway less than one.

Quote:
> And how is the hub for 10baseT operate ? Was the 10baseT designed
> for passive hub ? Or only active hub is possible ? For passive
> hub, I mean only passive electronics components are used, no
> electric power is necessary.

10baseT has always required an active hub. The only passive hub I've
heard of was for the old Arcnet network. 10base2 networks didn't
require a hub, instead using taps on a coaxial cable.

Quote:
> Why when you cascade the hubs, you need an MDI/MDIX switch for a
> specific port ? How many hubs can be chains at maximum ?

That depends on whether they're plain hubs or switched. The plain hubs
are treated the same as repeaters. The ports on a switched hubs are
essentially different networks, each of which follow the ethernet rules,
independent of the others.

Quote:
> Thanks a lot for your patient to read the long question.

> SK

--
Replies sent via e-mail to this address will be promptly ignored.

by **Sander Rooijen** » Sun, 08 Apr 2001 05:58:42

> > I asked this question because I'm studying the networking in details.

> > Over 10 years ago when I read books about networking, mostly they
> > cannot provide solid idea about networking. Until recent years, I
> > found some books describing bits and pieces of what is actually using.
> > Even you will not be able to successfully config your network by
> > reading one book, after you read several book plus some hands on,
> > you'll get it.

Experience doesn't come in books.
But they do help sometimes ;-)

Quote:
> > Still, if you want to know what is doing inside a repeater, bridge, ...
> > there are rare books telling you about this. They always refer such
> > devices to OSI 7 layer model and full stop. But its hardly for you
> > to know what would be the behaviour of the devices when you get to
> > work with them.

If you have a good understanding of the OSI model this can really help in
troubleshooting a network!
Especially when working with routers (layer 3) and bridges/switches (layer
2)
Also these books that refer to functions on a certain layer of the OSI model
will become clearer.

You might want to try a book for a CCNA course (Cisco Certified Network
Associate)
They usually will help you understand OSI as well as a basic understanding
of the functions of repeaters, bridges, switches and routers

Although less in-depth and a bit (lot) out of date there's also some useful
information on these subjects in Microsofts "Networking Essentials" (was
part of the NT4 MCSE track)

Quote:
> > My question is what is a repeater for 10base2 ? Because the repeater
> > is connected to 2 segments of co-axial, if there is signal from 1
> > segment, I can imagine the repeater will copy it to the other side.
> > but what if the other side also started to send ? If it copied,
> > thats OK coz eventualy both side should detect collision. Then it
> > must distinguish if the signal is send by itself or from the LAN
> > segment, that means the repeater should have a bit detection
> > rather than simply 2 amplifiers.

> Repeaters regenerate the signal, rather than just amplify it. Also,
> while they work in both directions, they do so in only one direction at
> a time.

Repeaters work only on layer 1 of the OSI model ;-)
Regenerate???
Some do...... but not all of them! Some just amplify!
The big problem with cable length at 185M it not the delay. It's the
attennuation.
Amplification solves this.
The disadvantage of amplification only is that noise picked up on one cable
segment will also be amplified and transmitted on the other segment.
regeneration prevents this.

If there are NIC's transmitting on both sides this will result in a
collission the same way it would as if they were on the same cable.
In one direction at a time? Yes but this is also the same as a cable!
Transmitting in 2 directions at the same time = collission.

A repeater doesn't make any forwarding decisions.
Whatever is one one segment will also be on the other segment. Both ways.
It's just like one network cable, only with a few S extra latency and more
available cable lenght and/or more than 30 taps on the network.

Quote:
> > Further, the 10base2 rely on the characteristic impedance of the
> > 50 ohm co-axial cable plus the maximum distance of 185M plus the
> > minimum 46 data bytes which sum up to 576 bits of stream to detect
> > the collision. (Anyone know how to interpret this calculation of
> > propagation delay ?) The specification said up to maximum of 5
> > segments with 4 repeater is possible. Is this collision scheme
> > base the calculation of propagation delay on full length for 185
> > X 5 plus repeater delay ?

> The segment length is determined by signal loss etc. The overall
> length, using multiple segments is determined by the minimum packet time
> i.e. smallest packet, travelling to the furthest point of the network.
> The minimum time is determined by the number of bits in the smallest
> allowable packet * 0.1 uS per bit. The maximum distance is determined
> by how far the signal will travel through the cable in that time. The
> speed of the signal = the speed of light * the velocity factor of the
> cable. The velocity factor is alway less than one.

There's a 5-4-3 rule.
On a 10Mbits network you can have 5 segments connected with 4 repeaters (or
hubs) with pc's connected to 3 of these network segments.
This does not mean that a network that doesn't adhere to these rules won't
work! It does mean that most hardware manufacturers will not support the use
of a network that doesn't.
If you read the manuals for the specific equipment you're using there should
be something in there about this (if it's a good brand).
The above calculation is missing the (important!) latency while going
through the hubs/repeaters.
This delay will usually be a lot more than the delay in any reasonable cable
lenght.

Have a look at this document:
ftp://ftp.compaq.com/pub/archives/networks/reptr-hr.pdf
is explains some of this stuff.

Quote:
> > And how is the hub for 10baseT operate ? Was the 10baseT designed
> > for passive hub ? Or only active hub is possible ? For passive
> > hub, I mean only passive electronics components are used, no
> > electric power is necessary.

> 10baseT has always required an active hub. The only passive hub I've
> heard of was for the old Arcnet network. 10base2 networks didn't
> require a hub, instead using taps on a coaxial cable.

10base2: As long as you don't have more than 30 taps on one segment (or
cable lenght of more than 185M of course)
Hubs: A hub really is like a repeater. Only with more ports. Sometimes
they're even called multiport repeaters (like in the Digital DEChub 90
series)

Quote:
> > Why when you cascade the hubs, you need an MDI/MDIX switch for a
> > specific port ? How many hubs can be chains at maximum ?

> That depends on whether they're plain hubs or switched. The plain hubs
> are treated the same as repeaters. The ports on a switched hubs are
> essentially different networks, each of which follow the ethernet rules,
> independent of the others.

Hubs do _not_ need an MDI/MDIX switch to be cascaded.
You can use a crossover cable if you cannot switch the port.
Sometimes hubs have one port with 2 connectors: one standard and one crossed
(use only one of them at the same time!)

The amount of hubs that can be cascaded depends on the latency (time it
takes when a signal arrives on an input until it's sent out).
Not all hubs do this at the same speed.
General rule of thumb: for 10 Mbits networks you can cascade a maximum of 4
hubs (like repeater segments. 5-4-3 rule)
for 100Mbits networks: don't cascade more than 2 hubs with a maximum of 5M
between the two hubs.
(these rules as said before do not apply for switches because then the ports
are not on the same collission domain)

Quote:
> > Thanks a lot for your patient to read the long question.

You're welcome :)

- Show quoted text -

Quote:
> > SK

by **WONG SAI-KE** » Sun, 08 Apr 2001 11:16:48

: You might want to try a book for a CCNA course (Cisco Certified Network
: Associate)
: They usually will help you understand OSI as well as a basic understanding
: of the functions of repeaters, bridges, switches and routers

In fact, I read CISCO network book and got better understanding of hub,
switch, router.

: Although less in-depth and a bit (lot) out of date there's also some
useful
: information on these subjects in Microsofts "Networking Essentials" (was
: part of the NT4 MCSE track)

I'm sorry I'm biased on M$ products (or documents), I tried to avoid using
or reading their product unless I have no choice.

:> > My question is what is a repeater for 10base2 ? Because the repeater
:> > is connected to 2 segments of co-axial, if there is signal from 1
:> > segment, I can imagine the repeater will copy it to the other side.
:> > but what if the other side also started to send ? If it copied,
:> > thats OK coz eventualy both side should detect collision. Then it
:> > must distinguish if the signal is send by itself or from the LAN
:> > segment, that means the repeater should have a bit detection
:> > rather than simply 2 amplifiers.

: Some do...... but not all of them! Some just amplify!
: The big problem with cable length at 185M it not the delay. It's the
: attennuation.

I can imagine.

: If there are NIC's transmitting on both sides this will result in a
: collission the same way it would as if they were on the same cable.
: In one direction at a time? Yes but this is also the same as a cable!
: Transmitting in 2 directions at the same time = collission.

: A repeater doesn't make any forwarding decisions.
: Whatever is one one segment will also be on the other segment. Both ways.
: It's just like one network cable, only with a few S extra latency and
: more available cable lenght and/or more than 30 taps on the network.

If you understand my question, you should not expect I was asking a stupid
question. May be I should explain the doubt in more details:

+----------+
segment A -------+ Repeater +--------- segment B
+----------+

If the circuit inside repeater is simply an bi-directional amplifier:

+-------------+
| |\ |
| +--+ >--+ |
| | |/ | |
segment A -----+--+ +--+-- segment B
| | /| | |
| +--< +--+ |
| \| |
| Repeater |
+-------------+

Any signal appears in segment A, after amplified to segment B will be
regenerated back to A. That is a circuit with positive feedback.
Doesn't matter whether its amplifier or regenerator, it suffers from
the same problem. My question is, there must be some means in the
repeater circuit to decide: not to re-amplify (re-generate) the signal
back to the source.

The only simple way I can think of is: signal appeared in whichever
side first should turn off the amplifier of the return path. If
collision occurred at the opposite side, then collision is also
injected into the original side.

Quote:> There's a 5-4-3 rule.
> On a 10Mbits network you can have 5 segments connected with 4 repeaters (or
> hubs) with pc's connected to 3 of these network segments.

Cannot understand why only connected to 3 of the segments ? Couldn't it
be 2 segments or 5 segments ?

Quote:> The above calculation is missing the (important!) latency while going
> through the hubs/repeaters.
> This delay will usually be a lot more than the delay in any reasonable cable
> lenght.

I think the manufacturer has to conform to standard for the
requirement of the latency. Otherwise, the 5 segments of 185M
under 10base2 could not be attend. Besides, I wouldn't imagine
much impact (from user's point of view) to the network performance
due to the propagation delay.

Quote:> Hubs: A hub really is like a repeater. Only with more ports. Sometimes
> they're even called multiport repeaters (like in the Digital DEChub 90
> series)

No, according to my previous question about the repeater for 10base2.
10base2 is a co-axial cable with only one signal line for both direction.
10baseT has 2 pairs, one Tx, one Rx. Therefore, 10baseT hub does not
need the intellegence of 10base2 repeater.

Quote:> Hubs do _not_ need an MDI/MDIX switch to be cascaded.
> You can use a crossover cable if you cannot switch the port.

Then, I understand. I overlooked the signal needs to be cross when
connecting up 2 hubs.

Quote:> The amount of hubs that can be cascaded depends on the latency (time it
> takes when a signal arrives on an input until it's sent out).
> Not all hubs do this at the same speed.
> General rule of thumb: for 10 Mbits networks you can cascade a maximum of 4
> hubs (like repeater segments. 5-4-3 rule)

+-----+ +-----+ +-----+ +-----+
---+ Hub +------+ Hub +------+ Hub +------+ Hub +---
+-----+ +-----+ +-----+ +-----+

General rule of thumb is OK, but don't forget other possibility

+-----+ +-----+
---+ Hub +--+---+ Hub +---
| | +-----+
| +--+
| | | +-----+
| | +---+ Hub +---
| | +-----+
| +--+
| | | +-----+
| | +---+ Hub +---
| | +-----+
| +--+
| | | +-----+
| | +---+ Hub +---
| | +-----+
| +--+
+-----+ | +-----+
+---+ Hub +---
+-----+

Quote:> for 100Mbits networks: don't cascade more than 2 hubs with a maximum of 5M
> between the two hubs.

Have no idea about how 100baseT is implement and its specifications.

SK

by **Sander Rooijen** » Sun, 08 Apr 2001 19:57:10

>.......

....... I'm sorry I'm biased on M$ products (or documents), I tried to avoid
using

Quote:> or reading their product unless I have no choice.

Well...
I'm a bit biased on M$ as well.... Which sensible person wouldn't be ;-)
But still... if you read carefully and critically, you can even learn from
their mistakes :-)
Don't believe everything you read.... M$ or otherwise (even if it it my
writing ;-).
Get a second opinion.

Quote:
> :> > My .....

........The only simple way I can think of is: signal appeared in whichever

Quote:> side first should turn off the amplifier of the return path. If
> collision occurred at the opposite side, then collision is also
> injected into the original side.

Of course... that's the only way this setup can work.
If a network interface is sending it's receiving part is not trying to
actually receive data.
It's used to detect collissions.
This is actually the same for 10baseT and 100baseT when used in half-duplex
mode.

Quote:> ........

...........> Cannot understand why only connected to 3 of the segments ?
Couldn't it

Quote:> be 2 segments or 5 segments ?

I agree with you here.
Okay.. I probably should have said a maximum of 3. Of course 2 will work.
To tell you the truth.. I don't know why either. I think if you use 4 or all
5 segments it will usually work just fine.
Somehow this is not within specifications though.

......
..... I think the manufacturer has to conform to standard for the

Quote:> requirement of the latency. Otherwise, the 5 segments of 185M
> under 10base2 could not be attend. Besides, I wouldn't imagine
> much impact (from user's point of view) to the network performance
> due to the propagation delay.

I think nowadays that'll ussually be the case. For 10Mbits it's not that
hard to meet the speed requirements.
for 100Mbits, In the documentation that comes with 3com hubs you can read
that for hub-to-hub connections you need "Class II" hubs (which of course
the 3coms are). I don't recall other manufacturers using this qualification
but maybe they do.

http://support.3com.com/infodeli/tools/hubs/off-con/3c16750/manual.a0...
14.htm#149782

This is not because of network performance due to propagation delay though.
It has to do with timing when collissions happen.
In general you don't want collissions to happen after more than 32 bits have
already been transmitted.
So the time it takes for a signal to hit the other end of the network and
return should be less than the time it takes to transmit 32 bits.
An increase in the number of collissions and their timing/handling (backoff
timers etc.) will have a lot more impact on performance than a few ms of
latency. And of course all of this is also very dependant on the network
load.

Quote:
> > Hubs: A hub really is like a repeater. Only with more ports. Sometimes
> > they're even called multiport repeaters (like in the Digital DECrepeater
90
> > series)

> No, according to my previous question about the repeater for 10base2.
> 10base2 is a co-axial cable with only one signal line for both direction.
> 10baseT has 2 pairs, one Tx, one Rx. Therefore, 10baseT hub does not
> need the intellegence of 10base2 repeater.

You're patially right.
In a 10baseT connection there are separate pairs for Tx and Rx.
But on a hub they cannnot be used at the same time though!
If a hub receives a signal on a port this is sent out of all ports except
the one that is sending it.
Therefore the return path on that connection can never be used unless it is
to indicate a collission.

For a full-duplex connection you need to be connected to a switch.
Then the Tx and Rx pairs _can_ be used independantly and at the same time

Quote:
> > Hubs .......

..... I overlooked the signal needs to be cross when

Quote:> connecting up 2 hubs.

Right.
All you need to do is make sure that the TX pair on one side connects to the
Rx on the other side and vice versa.
That's all.

Quote:
> > The amount .......

.........other possibility

Quote:
> +-----+ +-----+
> ---+ Hub +--+---+ Hub +---
> | | +-----+
> | +--+
> | | | +-----+
> | | +---+ Hub +---
> | | +-----+
> | +--+
> | | | +-----+
> | | +---+ Hub +---
> | | +-----+
> | +--+
> | | | +-----+
> | | +---+ Hub +---
> | | +-----+
> | +--+
> +-----+ | +-----+
> +---+ Hub +---
> +-----+

In this case you have to look at the maximum distance between any 2 hosts on
the network and apply the same rules to all possible connections.
This means that this setup is possible but quite limited. you cannot go down
4 levels from a hub connecting all branches because then the maximum lenght
path would go through 9 hubs where a maximum of 4 is specified

Quote:
> > for 100Mbits networks: don't cascade more than 2 hubs with a maximum of
5M
> > between the two hubs.

> Have no idea about how 100baseT is implement and its specifications.

In general it works the same, only faster.

You can apply the same rule:
In general you don't want collissions to happen after more than 32 bits have
already been transmitted.
So the time it takes for a signal to hit the other end of the network and
the collission signal to return should be less than the time it takes to
transmit 32 bits.

This means that on a 100baseT network you can not have as much latency as on
a 10baseT. Hence the limitation.

Regards,

Sander

by **WONG SAI-KE** » Tue, 10 Apr 2001 13:27:17

: You're patially right.
: In a 10baseT connection there are separate pairs for Tx and Rx.
: But on a hub they cannnot be used at the same time though!
: If a hub receives a signal on a port this is sent out of all ports except
: the one that is sending it.

I don't see any point for a hub to do this. Because hub has no
intellegence, I can assume it works only in half-duplex mode. It is
not necessary to add extra cost to stop resend to itself. Think about
that, if the circuit simply copy anything from any Tx to every Rx, its
simply and work fine for half-duplex. It makes no difference to
add circuit to resend it back to the sender cause the sender simply
ignore it, but the hub circuit will be costly.

:> Have no idea about how 100baseT is implement and its specifications.
: In general it works the same, only faster.

I don't think so. Even I haven't study the 100baseT, logically thinking,
it uses 4 pairs instead of 2 pairs. That means its not simply multiply
the serialization rate by 10. I remember some weeks ago, I read some
HP's document about her 100VG-AnyLAN, it says something like their is
better method in 100baseT to prevent collision or something similar.
But I'm sorry I could not find out the original message at the moment.

Regards,

SK

by **Sander Rooijen** » Tue, 10 Apr 2001 17:25:41

----- Original Message -----

Newsgroups: comp.os.linux.networking
Sent: Monday, April 09, 2001 6:27 AM
Subject: Re: What is the hardware inside repeater ?

> I don't see any point for a hub to do this. Because hub has no
> intellegence, I can assume it works only in half-duplex mode. It is
> not necessary to add extra cost to stop resend to itself. Think about
> that, if the circuit simply copy anything from any Tx to every Rx, its
> simply and work fine for half-duplex. It makes no difference to
> add circuit to resend it back to the sender cause the sender simply
> ignore it, but the hub circuit will be costly.

As it's only a few more transistors on the same chip. It's not that costly.
Compared to what's inside a lot of modern electronic equipment these chips
are not very complicated or fast.
You can even use the same existing port designs for 10base2 just with a
different transceiver
You're still only using one IC to do all the hard work.

If a network card is running half-duplex then data should not come in on the
Rx when it is sending itself.
The NIC has an internal loopback. This is used for collission detection.
Although it is using different cabling it is using the same media access
method as it would for 10base2.
(CSMA/CD: Carrier Sense Multiple Access with Collision Detection)
This is also why it is easy to integrate a 10base2 connection into a 10baseT
hub.
You might have seen those before.
Only in full-duplex mode a NIC is able to send and receive simultaneously.
It cannot run FD when connected to a hub.

> I don't think so. Even I haven't study the 100baseT, logically thinking,
> it uses 4 pairs instead of 2 pairs. That means its not simply multiply
> the serialization rate by 10. I remember some weeks ago, I read some
> HP's document about her 100VG-AnyLAN, it says something like their is
> better method in 100baseT to prevent collision or something similar.
> But I'm sorry I could not find out the original message at the moment.

Of course there is a little bit more to it than just increase the speed.
Just replacing the 20MHz crystal on a 10Mbits NIC with a 200MHz will not
make it a 100Mbits card.

100VG-AnyLAN is using 4 pairs and works on Cat.3 wiring.
Advantage: This makes it possible to run 100M over existing (older) wiring.
100VG-AnyLAN as far as I know has never been very popular.

100BaseT(x) "FastEthernet" is using the same 2 pairs as normal 10baseT but
needs Cat.5 wiring.
Disadvantage For old installations this might have the disadvantage that you
need new cabling.
But Cat.5 has been the standard for several years already. So has
FastEthernet.
Advantage: This makes it relatively easy (and therefore cheap) to create
things like 10/100 NIC's because the 2 speeds can share a lot of the
circuitry on the card.
It provides an easy upgrade path from 10 to 100. Even running simultaneously
in the same network with 10/100 hubs or switches.

I assumed you were talking about FastEthernet and not VG-AnyLAN.
Maybe I shouldn't have but it is the standard nowadays.
There are many other 100Mbits technologies available. A few examples:
There is a Token Ring 100Mbits standard now.
Also FDDI has been doing 100Mbits for many years, even before the
fastethernet standards and also over copper wiring.
Of course you can use all of them to create a "faster than 10" connection.
All have their own advantages and disadvantages.

FastEthernet is the one that is most similar to 10Mbits Ethernet and
therefore the only one where what I've said applies.
You cannot plug a VG-AnyLAN into the same (cheap) hub (well... okay.... yes
you can... but don't expect it to work ;-).

Regards,

Sander

by **Donald Beck** » Wed, 11 Apr 2001 00:13:36

>: You're patially right.
>: In a 10baseT connection there are separate pairs for Tx and Rx.
>: But on a hub they cannnot be used at the same time though!
>: If a hub receives a signal on a port this is sent out of all ports except
>: the one that is sending it.

>I don't see any point for a hub to do this. Because hub has no
>intellegence, I can assume it works only in half-duplex mode. It is
>not necessary to add extra cost to stop resend to itself.

Yes, it is necessary, otherwise the adapter wouldn't be able to tell if
there is a collision or it is just seeing its own transmission.

10base2 and 10base5 detect collisions by measuring the voltage on the
coax. This isn't possible with long runs of twisted pair.

Quote:>:> Have no idea about how 100baseT is implement and its specifications.
>: In general it works the same, only faster.

>I don't think so. Even I haven't study the 100baseT, logically thinking,
>it uses 4 pairs instead of 2 pairs.

No, it uses only 2 pair. The other 2 pair in the Cat5 cable are usually
terminated to reduce coupled noise.

Quote:>That means its not simply multiply the serialization rate by 10.

The encoding is substantially different. 10baseT encodes one bit at a
time. 100baseTx encodes 4 bits into each symbol, and sends 25M symbols
per second.

Quote:>I remember some weeks ago, I read some
>HP's document about her 100VG-AnyLAN, it says something like their is
>better method in 100baseT to prevent collision or something similar.

The marketing spin at the time was so extreme I don't think that you
will find a _fair_ comparison between 100VG and 100baseTx.
100VG did use more pairs, and did have some some encoding advantages.
However 100VG was more complicated and cost more to implement.
There were good reasons that 100baseTx won.

--

Scyld Computing Corporation http://www.scyld.com
410 Severn Ave. Suite 210 Beowulf Clusters / Linux Installations
Annapolis MD 21403