Switch Question

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

Hello,

I have a hypothetical question with practical implications: Two
computers, 'A' and 'B', are plugged into a 100baseTX switch. Computer
'A' has a 100baseTX interface and computer 'B' has a 10baseT card.
When 'A' is talking to 'B' through the switch, does it send its frames
at the slower 10 Mbit/sec rate, or does it send to the switch at the
full 100 Mbit/sec and the switch shift it out slower to 'B'?

- Andy Ball

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

Andy Ball wrote:

>
> Hello,
>
> I have a hypothetical question with practical implications: Two
> computers, 'A' and 'B', are plugged into a 100baseTX switch. Computer
> 'A' has a 100baseTX interface and computer 'B' has a 10baseT card.
> When 'A' is talking to 'B' through the switch, does it send its frames
> at the slower 10 Mbit/sec rate, or does it send to the switch at the
> full 100 Mbit/sec and the switch shift it out slower to 'B'?

"A" will send at the full rate. If the switch doesn't use flow control, the
buffers will eventually fill and frames will be lost. However, the TCP
protocol also has it's own flow control mechanisms, that cause it to back
off, when switch congestion causes delays or lost frames. This will reduce
the amount of data that "A" tries to send through the switch.

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

James Knott wrote:

> Andy Ball wrote:
>
>>
>> Hello,
>>
>> I have a hypothetical question with practical implications: Two
>> computers, 'A' and 'B', are plugged into a 100baseTX switch. Computer
>> 'A' has a 100baseTX interface and computer 'B' has a 10baseT card.
>> When 'A' is talking to 'B' through the switch, does it send its frames
>> at the slower 10 Mbit/sec rate, or does it send to the switch at the
>> full 100 Mbit/sec and the switch shift it out slower to 'B'?
>
> "A" will send at the full rate. If the switch doesn't use flow control,
> the
> buffers will eventually fill and frames will be lost. However, the TCP
> protocol also has it's own flow control mechanisms, that cause it to back
> off, when switch congestion causes delays or lost frames. This will
> reduce the amount of data that "A" tries to send through the switch.

Forgot to mention: look up IEEE 802.3x flow control. This flow control
mechanism uses a "pause" MAC control frame to tell another device to stop
sending for a specified period of time. However, flow control is optional,
so not all switches use it.

 

[Stephen]

Archived from groups: comp.dcom.lans.ethernet (More info?)

"Andy Ball" <null@not.valid> wrote in message
news:%v1Ee.1999$Uk3.956@newsread1.news.pas.earthlink.net...
>
> Hello,
>
> I have a hypothetical question with practical implications: Two
> computers, 'A' and 'B', are plugged into a 100baseTX switch. Computer
> 'A' has a 100baseTX interface and computer 'B' has a 10baseT card.
> When 'A' is talking to 'B' through the switch, does it send its frames
> at the slower 10 Mbit/sec rate, or does it send to the switch at the
> full 100 Mbit/sec and the switch shift it out slower to 'B'?

it is a switch, so each port operates at the local line rate. a device cant
"see" the line rate of other devices whether they are on the same switch or
across the world.

the switch will have some internal buffering, so there is some Q capacity
between the 2 ports.

either a higher layer protocol will control the amount of data sent so that
the connection operates with a tolerable average loss rate (e.g. TCP), or
the traffic stream is uncontrolled, so the Q will overflow and the switch
will throw away some packets.
>
> - Andy Ball
--
Regards

Stephen Hope - return address needs fewer xxs

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

James Knott wrote:

> Forgot to mention: look up IEEE 802.3x flow control. This flow control
> mechanism uses a "pause" MAC control frame to tell another device to stop
> sending for a specified period of time. However, flow control is optional,
> so not all switches use it.

It's optional at 100 Mbps but mandatory at 1 Gbps and above.

However, most folks caution against enabling 802.3x for any ports
other than those connected directly to an end-station because
it causes head of line (HOL) blocking.

Anoop

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

Hello Stephen,

S> it is a switch, so each port operates at the local line
> rate. a device cant "see" the line rate of other
> devices whether they are on the same switch or across
> the world. the switch will have some internal
> buffering, so there is some Q capacity between the 2
> ports.

That's good to hear. Hopefully it means that a server with
a 100baseTX port can service a few workstations that each
have 10baseT ports without one workstation being able to
saturate the server's connection to the switch. Is it safe
to assume that the same would apply if the server had 1000-
baseCX and the workstations had 100baseTX, or the server had
10 gigabit and the workstations had 1000baseCX ports (given
wire-speed switches) ?

- Andy.

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

Hello James,

JK> Forgot to mention: look up IEEE 802.3x flow control.
> This flow control mechanism uses a "pause" MAC control
> frame to tell another device to stop sending for a
> specified period of time. However, flow control is
> optional, so not all switches use it.

Thanks, that's useful information. Something to look for on
switch data sheets too.

- Andy.

 

[Stephen]

Archived from groups: comp.dcom.lans.ethernet (More info?)

"Andy Ball" <null@not.valid> wrote in message
news:%2tFe.5154$Uk3.2653@newsread1.news.pas.earthlink.net...
>
> Hello Stephen,
>
> S> it is a switch, so each port operates at the local line
> > rate. a device cant "see" the line rate of other
> > devices whether they are on the same switch or across
> > the world. the switch will have some internal
> > buffering, so there is some Q capacity between the 2
> > ports.
>
> That's good to hear. Hopefully it means that a server with
> a 100baseTX port can service a few workstations that each
> have 10baseT ports without one workstation being able to
> saturate the server's connection to the switch. Is it safe
> to assume that the same would apply if the server had 1000-
> baseCX and the workstations had 100baseTX, or the server had
> 10 gigabit and the workstations had 1000baseCX ports (given
> wire-speed switches) ?

yes - with the very big assumption that the network is the bottleneck and
not the server

in practice 10/100 ports are the minimum available on modern Ethernet
switches - and 10/100/1000 ports are now standard on some low to mid range
laptops....
>
> - Andy.
--
Regards

Stephen Hope - return address needs fewer xxs