Without doing any research and only depending on the seat of my pants, here
is what I believe.
In a cut-through environment, the switch will only buffer the necessary
portion of the frame that is necessary for it to know where to switch the
frame to. It doesn't care if the frame is a jumbo or not, a collision or
not. Therefore if a packet is received destined for the broadcast address
and the header is in tact and looks completely correct, yet the packet is
hacked all to hell because of a collision beyond the header information,
this collision will be broadcast to all devices.
In a store-and-forward environment, the switch buffers the entire packet and
handles error conditions so that end-systems do not have to be interrupted
in order to process erroneous packets.
I believe that on a number of switches, they can be configured as
cut-through with some extra checking. Should a port be receiving a large
number of erroneous packets, they will reconfigure themselves on the fly to
store-and-forward in order to guarantee the highest throughput possible.
These are just my quick thoughts, I'm sure a little research would lead to a
more fully-qualified answer. But, you get what you pay for.
David C Prall [EMAIL PROTECTED] http://dcp.dcptech.com
----- Original Message -----
From: "Priscilla Oppenheimer" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Monday, June 19, 2000 6:13 PM
Subject: RE: collision on cut-through switch
> The 1900 and 2820 Catalyst switches do cut-through. I did find a good
white
> paper at:
>
>
http://www.cisco.com/warp/public/cc/cisco/mkt/switch/cat/c1928/tech/nwgen_wp
.htm
>
> But it never comes right out and says what happens upon a collision when
> forwarding. I can guess based on the architecture, but I'm looking for
some
> more definitive answers to prove my point with the other instructor. ;-)
>
> Priscilla
>
>
>
> At 04:30 PM 6/19/00, Daniel Cotts wrote:
> >Are any current Cisco switches using cut through? The 5000s use Store and
> >Forward. The old Kalpana switches aka Catalyst 3000 used cut through.
> >Looking at an old EtherSwitch PRO16 manual (same same cat3k) it mentions
> >on-board buffering. "If the destination port is receiving a packet from
> >another EtherSwitch PRO16 port or if the output segment is busy, the
> >EtherSwitch PRO16 stores the packet in one of its on-board buffers. Each
> >EtherSwitch PRO16 buffer can hold up to 384 packets in each direction
> >(incoming and outgoing). This helps balance throughput when networks are
> >operating near peak load and more than one packet may be directed to the
> >same port at the same time."
> >Hope this helps.
> > > -----Original Message-----
> > > From: Priscilla Oppenheimer [mailto:[EMAIL PROTECTED]]
> > > Sent: Monday, June 19, 2000 3:47 PM
> > > To: [EMAIL PROTECTED]
> > > Subject: collision on cut-through switch
> > >
> > >
> > > Hi Group Study,
> > >
> > > I got into a discussion with a knowledgeable Sniffer
> > > instructor recently.
> > > When he teaches cut-through-switching theory, he warns his
> > > students that a
> > > cut-through switch does not really isolate collision domains.
> > > Consider this
> > > example:
> > >
> > > * The switch is receiving a frame from port 1 destined for a
> > > station out
> > > port 2.
> > > * The switch recognizes the destination address and starts
> > > forwarding the
> > > frame to port 2 ASAP.
> > > * There is a collision on port 2. (It's a shared and/or
> > > half-duplex Ethernet.)
> > >
> > > According to the instructor, the Switch sends a jam signal
> > > back to port 1
> > > to let the initial sender know that the frame experienced a
> > > collision. This
> > > allows the sender to retransmit.
> > >
> > > If you read some of the books on switching, you would think
> > > that this is
> > > true. The books make it sound like the frame is passing
> > > through the switch
> > > and disappearing out the destination port as soon as the destination
> > > address is recognized.
> > >
> > > I don't think the Sniffer instructor's conclusion is true, however. I
> > > believe that a Cisco cut-through switch buffers the frame and
> > > hence has the
> > > ability to retransmit. There is no requirement to send a jam to the
> > > original sender because port 2 in our example retransmits
> > > after sensing the
> > > collision.
> > >
> > > I believe that Cisco switches store frames, even when doing
> > > cut-through,
> > > whereas the instructor assumed that the frame has passed
> > > through and out
> > > the port and is no longer available for retransmission by the switch.
> > >
> > > Cisco positions cut-through as reducing delay, not reducing
> > > the need for
> > > buffering, so I'm contending that I'm right.
> > >
> > > Who do you think is right? Can you point me to any white
> > > papers that would
> > > prove who is right?
> > >
> > > Thanks
> > >
> > > Priscilla
> > >
> > > ________________________
> > >
> > > Priscilla Oppenheimer
> > > http://www.priscilla.com
> > >
>
>
> ________________________
>
> Priscilla Oppenheimer
> http://www.priscilla.com
>
> ___________________________________
> UPDATED Posting Guidelines: http://www.groupstudy.com/list/guide.html
> FAQ, list archives, and subscription info: http://www.groupstudy.com
> Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
>
___________________________________
UPDATED Posting Guidelines: http://www.groupstudy.com/list/guide.html
FAQ, list archives, and subscription info: http://www.groupstudy.com
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
