I never said it didn't start forwarding when it got the destination address, I said it read both (at least in some switches)
The question had to do with address learning. When a frame is recieved on an interface, it takes the address and caches it and the port it came in on (if it doesn't already have it). This doesn't mean the switch cannot start forwarding the frame as soon as it has the destination address.
Also, I assume that the techniques will differ from vendor to vendor.
Intel says this about cut-through and its switch:
Much better performance is achieved with either cut-through forwarding or fragment-free forwarding. In the case of cut-through switching, only the first 14 bytes are buffered. As soon as the switch has read the destination and source addresses and the type field, the frame is forwarded on the destination port. (Actually, the forwarding decision could be made as soon as the destination address is read. But doing so would prevent filtering on the source address and protocol content of the frame).
The main difference between cut-through and store and forward is that cut through sends the data as soon as it has then information it needs to send it (the destination address) and store and forward gets it all first.
You have to differentiate "forwarding" from "address learning".
Very comprehensive tfs.
The selling point for cut through is of course the speed, while vendors of store-and-forward devices will tell you that their approach eliminates the forwarding of all invalid or corrupt packets. I did a head-to-head once to test all this and to be honest I couldn't see much difference on a 'normal' network. I think the speed argument may get the edge under heavy load and latency-crucial applications while the cleanliness argument might win if your network is full of crud. Take your pick!
Re: Cut-thru - Addr Learning
13 years 7 months ago #6700
well personally I think cut through is lacking in the security realm as well, because when you think about it, if you have a MAC based filter set up on a cut through switch, it would need be cached before being checked by the interfaces filter, defeating the purpose of the speed augmentations of cut through. Personally, with the availability and advancement of cheap fast media topologies, advantages of store and fowards far outweigh the disadvantages.