Previous | Next --- Slide 7 of 46
Back to Lecture Thumbnails
username

A way communication latency can be decreased is if the developer knows the type of communication that has to occur between separate processes, the data can be formatted in such a way to increase the speed of the communication. For example, in this case both students could have started walking towards each other WHILE doing their individual additions, thus by the time they were done with their individual tasks, they would be in a great position to communicate fast.

rootB

username mentioned that students can walk while doing addition but it's not necessarily doable in the context of processes processing tasks. The communication and addition could be two different tasks and it needs two different processes to handle these two kinds of task simultaneously but you can't take communication out of addition, because they have dependencies. Therefore I think these two tasks have to happen sequentially and you can't do communication while processing tasks. Increasing the speed of communication might be a great idea; I just thought the example isn't very appropriate.

mario

I partially agree with rootB that there are dependencies with the addition and communication between the different people. Something that I got from the demos is that communication is difficult to optimize for despite the amount of processors used to reduce the task load. This was seen when we had demo 3 and the students at the end of the rows took a while a to add the numbers since they were so large.

paramecinm

@ username Communication between processors may be more like making a phone call rather than one walking towards another. Thus I think the way you described may be not appropriate.