Alejandro Acosta Alamo wrote: > > .....I found over 3000 out of 15.000 packets that were less > than 64 bytes > longer..
First, I would check how your protocol analyzer counts. An application can send a packet that is less than 64 bytes. The Ethernet driver must pad the frame out to 64 bytes. It's possible your analyzer doesn't count the padding and is giving you a length that just counts the actual data. If that's not the explanation, then you have an unhealthy network. The recipient of an Ethernet frame that is less than 64 bytes will just drop the frame. Why waste bandwidth on this junk? On a shared network, frames that are less than 64 bytes (fragments or runts) can occur due to collisions. Some collisions are normal on a shared network. But 20% is too high. Runts result when a station starts sending, gets through the preamble and into the actual frame, but not past 64 bytes, and notices that another station is sending. The transmitters send a short jam, and then stop sending. They backoff and wait a random amount of time before retransmitting. The result is that they transmitted runts (frames less than 64 bytes). On a point-to-point network (supposedly not shared), runts result from a duplex mismatch. One side thinks the link is full duplex and sends whenever it wants. The other side is set for half duplex and thinks that receiving while it's sending is a collision, stops sending, backsoff, and leaves behind a runt. Priscilla > > ----- Original Message ----- > From: "Alejandro Acosta Alamo" > To: "Priscilla Oppenheimer" ; > > Sent: Tuesday, July 09, 2002 2:10 PM > Subject: Re: Cut-through vs Store & Forward [7:48316] > > > > Hello again, > > Priscilla, you have said that an ethernet frame must be at > least 64 > bytes, > > right?. I have just placed an sniffer on my LAN and I found > over 3000 out > of > > 15.000 packets. Does this mean that 20% of those packets are > illegal? > > > > Thanks > > > > Alejandro Acosta > > > > ----- Original Message ----- > > From: "Priscilla Oppenheimer" > > To: > > Sent: Monday, July 08, 2002 1:08 PM > > Subject: RE: Cut-through vs Store & Forward [7:48316] > > > > > > > Alejandro Acosta Alamo wrote: > > > > > > > > Hello, > > > > I understand the differences between Cut-through and > Store & > > > > Forward. My > > > > question is: How do you decide with method to use?, in > whch > > > > situation have > > > > you change the switching method?. > > > > > > > > Thanks > > > > > > > > Alejandro Acosta > > > > > > > > > > > A lot of switches support only one method, so you don't > have a choice. > If > > > you do have a choice, the decision is based on the number > of errors on > > your > > > network. Cut-through doesn't do any error checking and in > fact forwards > > > frames that have a bad CRC or are too short. Ethernet says > that frames > > must > > > be at least 64 bytes. Anything less is considered a > fragment and is > > illegal. > > > Cut-through forwards fragments that have an entire > destination address > > that > > > can be looked up to get a port number. > > > > > > If your switch connects many shared networks, then CRC > errors and > > fragments > > > due to collisions are normal. But why waste bandwidth > forwarding these > to > > > other ports on the LAN? In this case, you might want to go > with > > > store-and-forward which does not forward errored frames or > fragments. > > > > > > If your switch connects single devices all using > full-duplex, then it's > > > unlikely that you are experiencing many CRC or fragments. > So, > cut-through > > > makes the most sense. > > > > > > > > > Priscilla Oppenheimer > > > http://www.priscilla.com > > Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=48448&t=48316 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
