A switch port definitely delineates collision domains. Stations reachable
on different ports don't share the media and are not concerned with each
other's collisions. That answers part of the question. The link that I
found before (and can't find now) confirmed that this is true even with
cut-through switching. The switch caches the frame even when doing
cut-through and if it encounters a collision while outputting, it backs off
and retransmits. It does not send a collision enforcement jam to the
original sender.
If the Gigabit Ethernet link between the switches is set to full-duplex,
the switches can send to and receive from either at the same time. But they
can still only output one frame at a time. The Ethernet port outputs bits
serially. It's not a parallel port.
On the other hand, at 1,000 Mbps those bits are whipping out so fast, that
the question is pretty irrelevant. The "serialization delay" isn't
significant.
On the wire, the bits are whipping by at about 2/3 the speed of light in a
vacuum. (Yes, even fiber-optic cabling still provides a speed that is a
fraction of speed of light in a vacuum.) Theoretically on a long cable,
there could be more than one frame on the wire at a time. The math is
making my head hurt, but this is logical, if possibly not "real-world."
Ten stations outputting 100 Mbps all destined to the other switch will
cause the switch to queue packets. But stations don't output that fast.
They wait for ACKs coming back in the other direction, etc. Also, they
probably sometimes talk to other stations on their same switch. All your
traffic probably won't go across the Gigabit Ethernet link. How much
queuing will occur is hard to predict, but could be estimated with a good
analysis of traffic patterns. Or you could throw money at the problem and
group multiple Gigabit Ethernet ports together in an EtherChannel.
I think this was a good question (not elementary after all!) Answers could
vary from my off-the-cuff answer to a scientific and mathematical answer
that gets into the different types of delay (serialization, propagation,
queuing) and switch architectures.
Priscilla
At 11:42 AM 5/30/01, Allen May wrote:
>I believe it was Priscilla that found a link explaining all that in a white
>paper a couple months ago. I believe it backed up what you're saying...but
>I've slept (occasionally) since then.
>
>----- Original Message -----
>From: "Chuck Larrieu"
>To:
>Sent: Wednesday, May 30, 2001 10:14 AM
>Subject: RE: elementary? [7:6359]
>
>
> > Hhhmmmm....... Not so sure this is exactly right......
> >
> > With full duplex, you have effectively created two "directions" --- there
> > and back.
> >
> > I believe it is accurate to say that only one packet can be on the wire
>per
> > direction at one time.
> >
> > I can send to you at the same time you are sending to me. But Someone
else
> > can not send to you at the time my packet is on the wire.
> >
> > Correct me if I'm wrong.
> >
> > Chuck
> >
> > -----Original Message-----
> > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] On Behalf Of
> > Peter I. Slow
> > Sent: Wednesday, May 30, 2001 7:40 AM
> > To: [EMAIL PROTECTED]
> > Subject: Re: elementary? [7:6359]
> >
> > Noooooooo.
> > nononononono.
> > CSMA/CD only gets used when you are not in full duplex. (/me ducks) ( i
> > have NEVER seen a full-dup. hub) meaning that if i am using a switch
>capable
> > of full duplex (as most are) ..conversations, every station can transmit
>as
> > much as they want. this is what differentiates between a hub and a
switch.
> > (but not the only thing)
> > you are correct in that a 100 meg HUB with a gig uplink could never
fully
> > utilize the link, but the case is completly different with a switch.
> >
> >
> >
> > ----- Original Message -----
> > From: "Vijay Ramcharan"
> > To:
> > Sent: Wednesday, May 30, 2001 9:54 AM
> > Subject: RE: elementary? [7:6359]
> >
> >
> > > Thanks everyone for their replies. As I now understand it, the 1Gb
> > > uplink just moves data faster than... say, a 100Mb uplink. Correct?
> > > Conversations between hosts on each switch still take place one at a
> > > time, thereby obeying Ethernet rules of one station transmitting at a
> > > time. Correct?
> > > Okay my next question. Is there any point at which this 1Gb uplink can
> > > become saturated, since it's only handling station to station sessions-
> > > one at a time.
> > > If a number of stations on each switch were doing large file transfers
> > > to each other via the uplink, would there be some point at which the
> > > uplink would be maxed out- in terms of bandwidth? Or is the only
> > > limiting factor, the workstations inability to pump data out fast
enough
> > > to max out the uplink when they're only running 100Mb?
> > >
> > > I'm thinking that it's really not possible to max out a 1Gb uplink when
> > > stations are only running 100Mb. If this is correct then I lay this
> > > question to rest.
> > >
> > > Thanks.
> > >
> > > Vijay Ramcharan
> > >
> > >
> > > -----Original Message-----
> > > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] On Behalf Of
> > > Vijay Ramcharan
> > > Sent: Wednesday, May 30, 2001 12:06 AM
> > > To: [EMAIL PROTECTED]
> > > Subject: elementary? [7:6359]
> > >
> > >
> > > Forgive me if this sounds a little bit basic but this is what happens
> > > when you rush into things without understanding the fundamentals.
> > > Suppose a 24 port 100Mbit switch called A is uplinked to another 24
port
> > > 100Mb switch called B via a 1Gb connnection. Suppose hosts D through N
> > > are on switch A and hosts M through X are on Switch B. Would
> > > conversations between the hosts from Switch A to Switch B occur one at
a
> > > time or are multiple conversations multiplexed over the 1Gb uplink?
> > >
> > > I'm just trying to find out if and how that 1Gb uplink is used up.
> > > Thanks in advance. I'd put TIA but I hate those little acronyms. No
> > > flames please.
> > >
> > > Vijay Ramcharan
> > > FAQ, list archives, and subscription info:
> > > http://www.groupstudy.com/list/cisco.html
> > > Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
> > > FAQ, list archives, and subscription info:
> > http://www.groupstudy.com/list/cisco.html
> > > Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
> > FAQ, list archives, and subscription info:
> > http://www.groupstudy.com/list/cisco.html
> > Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
> > FAQ, list archives, and subscription info:
>http://www.groupstudy.com/list/cisco.html
> > Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
>FAQ, list archives, and subscription info:
>http://www.groupstudy.com/list/cisco.html
>Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
________________________
Priscilla Oppenheimer
http://www.priscilla.com
Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=6526&t=6359
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
