I don't understand, it says up until now packets were the granularity of transfer, but for cut-through routing weren't we kind of allowing packets to be broken up into smaller pieces?

I think the difference is that in cut-through, whenever a packet is blocked the whole packet gets buffered at the switch, but with flits you only have 1 flit buffered at the switch.

Just for confirmation:

In "cut through" flow control, routing information is just stored in the header.

Is the following statement true or not:

In "wormhole" flow control, every flit contains the routing information. So the body flits can route themselves, even without the head flit?

@muyangya I think this statement is not true. Only the head flit contains the routing information and the body flits just follow the head flit.

@muyanya I agree with cgjdemo. I think there's some control logic in each switch so that it keeps records of the head flit and knows where the body flits should go.

I think the difference between "cut through" and "wormhole" is: in cut through, the body flits and the tail flit keep moving until then reached the head flit even if the head is blocked.

@cgjdemo @MediumPeps Thanks, that makes a lot of sense. In cut through approach, the router need to have enough space(buffer) to hold the whole package. However, in wormhole approach, we just need to have enough space for a single split in the routers. This helps a lot to reduce the cost.