Could someone give me a technical reason why, when I connected a cable modem
and three computers to a hub, it was faster than connecting a cable modem
and three computers with a switch?

I know that routers are the solution, but I am in a debate and am drawing on
my past experience. For example, when using a packet sniffer, you want to
be a part of the same collision an broadcast domain. You can do that with a
hub, but you can't with a switch.

It must be the same principle when you only have 2 or 3 computers on the
same network and collisions are not a problem. Am I right and if I am,
could someone explain why?

TIA

f/fgeorge wrote:
introduce some amount of throughput delay into the data transfer from
input port to output port. The delay can be more or less depending on
which class of switch; "store and forward", or "cut-through" is used.

SF is worst because the complete incoming frame has to be received and
"stored" in the switch before it is allowed to start being sent out the
output port. This is primarily for error checking purposes.

CT is faster because data starts being output as soon as switch has
enough data to know which output port to send it to.

The most common mis-understanding about switches vs hubs is that
switches are not faster in all applications. The worst application is
where several PCs are trying to access the same one target ( ie DSL
modem) (many to one network). Hubs are actually faster here because they
have no delay as above.

Switches work best in applications where the PCs can access many targets
in the local network(ie, DSL, file servers, networked printers, PC-PC
transfer, etc) (any to any network)

--reed

In article <2jPkf.1311$Of1.142@newsread3.news.pas.earthlink.n et>,
Reed <reedh@rmi.net> wrote:

Does that mean that switches which support speed switching (e.g. 10/100
Mbps) will always store and forward when moving data between a 10 Mbps
port and a 100 Mbps port?

--
Tom Stiller

PGP fingerprint = 5108 DDB2 9761 EDE5 E7E3
7BDA 71ED 6496 99C0 C7CF

Tom Stiller wrote:

Even cut through switches have to store data, while waiting for previous
data to pass. So the first packet will start as cut through, but more and
more of it will be stored. Any additional packets behind it, will be
stored until the wire is free. However, TCP has mechanisms to detect such
congestion and back off the transmission rate, to reduce that congestion.
Also, some switches can cause a source to slow down.

James Knott wrote:
On the other hand, with a hub there is a potential for more collisions which
will result in delays and re-transmissions. So while there may be an initial
(very minute) hit to through-put when replacing a hub with a switch, as more
devices are added, the switch becomes more and more efficient.

But if we're talking about an internet connection using a cablemodem (as
would be on-topic for this group), this is largely insignificant. It's like
comparing a 0.1% upgrade on a highway vs. a stop sign further up the road.
Hardly worth the time it takes to consider.

--
Warren H.

==========
Disclaimer: My views reflect those of myself, and not my
employer, my friends, nor (as she often tells me) my wife.
Any resemblance to the views of anybody living or dead is
coincidental. No animals were hurt in the writing of this
response -- unless you count my dog who desperately wants
to go outside now.

This fall, vacuum up your leaves instead of raking:
http://www.holzemville.com/mall/blac...r/blowers.html

Tom Stiller wrote:
fail when the 10M port is the input and 100M is the output (causing an
under-run condition).

Also, it appears some posters still miss the point about sw vs hub being
application dependent. Switches were developed to help those local
networks that had a need for any to any connectivity. When 10 (or any
number more than 1) users on PCs are trying to access the same
cablemodem, a switch is no help, and may actually hurt, as the OP may
have learned the hard way.

--reed

From: "Reed" <reedh@rmi.net>


| IIRC, all 10/100 switches are store and forward, since cut-through would
| fail when the 10M port is the input and 100M is the output (causing an
| under-run condition).
|
| Also, it appears some posters still miss the point about sw vs hub being
| application dependent. Switches were developed to help those local
| networks that had a need for any to any connectivity. When 10 (or any
| number more than 1) users on PCs are trying to access the same
| cablemodem, a switch is no help, and may actually hurt, as the OP may
| have learned the hard way.
|
| --reed

It depends on the POV. The E-switch on a Cable/DSL Router is also being used to share local
area networking such as IPX/SPX, SMB or NetBIOS over IP. There can be 253 nodes on such a
SOHO LAN. The collision domain isolation of an E-switch will help on the LAN side very
well. Additionally, since an E-Switch can send and receive at the same time and so can DSL
or Cable Internet connections, there is that benefit as well.

The problem is that cheap SHO Router or cheap SOHO standalone E-switches use cheap chip-sets
with higher latency than their enterprise.corporate counterparts. The trade-off is lower
price for higher latency.

--
Dave
http://www.claymania.com/removal-trojan-adware.html
http://www.ik-cs.com/got-a-virus.htm

On Sun, 04 Dec 2005 23:11:47 GMT, "RobW" <robwl-no-spam@gmail.com>
wrote:

terribly slow speeds compared to a connection on the Hub that was
autonegotiating fine.

I run into this all the time. The first thing I do with any ethernet
connection of any kind is to force the devices to full duplex/100
Mbps. This will eliminate any device from guessing incorrectly during
the N-way negotiation.

With the speeds involved with Home Broadband the various types of
switching methods will have little to do with latency, assuming the
switch didn't come out of a cracker jack box...