Question What's the difference between using flits and decreasing packet size to the size of the flit? Is it because the body and tail flits don't have all the information to travel by themselves and can only follow the head flit? In that case, is the reason why flits are preferable because there's less total data to transfer? (Less space allocated for head/tails?)
This comment was marked helpful 0 times.
@Xelblade I think the difference between the 2 is a matter of abstraction vs implementation. You can think of packets as individual pieces of information a source node wants to send to a destination. How that data eventually reaches the destination (flits vs whole packets) doesn't really matter to the source as long as its message arrives and it doesn't really matter to the destination as long as the destination receives the entire packet.
The abstraction: sending individual piece of information from source to destination.
The implementation: flits.
This comment was marked helpful 1 times.
Of all the caterpillar pictures that could have been used, you ended up choosing one of the truly terrifying and rare carnivorous caterpillars. This comment isn't useful, but I thought I'd share.
Here's how I remember packet-based routing. Imagine the packet as a tour group (and the people are the data).
Store-and-forward is like moving from location A to location B on your tour, but you only move from location B to C once everyone in the tour group (packet) gets from A to B. Cut-through flow is similar, but instead of waiting for everyone to get from A to B, those who get to B are immediately funneled to C. Wormhole flow cuts the large tour group (packet) into mini-groups (flits) that are numbered. Each mini-group waits for everyone in their mini-group to arrive at location B and are then immediately taken to location C. More than one mini-group can be at a location, but no higher numbered group can be at a later location than any lower numbered group. Also, if group #1 (the head) stops, all the other groups can't move.