It might seem like latency and bandwidth are almost the same; you might think bandwidth is 1/latency (the # of requests you can serve is inversely proportional to the amount of time per request)
But in fact, they are totally orthogonal, because you can start the next request before the first one completes. To take the example from class, imagine students running down a hallway. The latency is the length of the hallway (how long it takes one student to complete). The bandwidth is the width of the hallway (how many students can run at once). If you imagine a very large crowd of students, you can see that (hallway width) of them are crossing the finish line each second, no matter how long the hallway is. So bandwidth is independent of latency!
This comment was marked helpful 1 times.
rutwikparikh
The definition of cost here is the effect of latency, bandwidth or other potential difficulties on our program at runtime. For example, a memory reference might have high latency but it may be improved upon by multi-threading and as a result, there would be no real cost on the program.
It might seem like latency and bandwidth are almost the same; you might think bandwidth is 1/latency (the # of requests you can serve is inversely proportional to the amount of time per request)
But in fact, they are totally orthogonal, because you can start the next request before the first one completes. To take the example from class, imagine students running down a hallway. The latency is the length of the hallway (how long it takes one student to complete). The bandwidth is the width of the hallway (how many students can run at once). If you imagine a very large crowd of students, you can see that (hallway width) of them are crossing the finish line each second, no matter how long the hallway is. So bandwidth is independent of latency!
This comment was marked helpful 1 times.
The definition of cost here is the effect of latency, bandwidth or other potential difficulties on our program at runtime. For example, a memory reference might have high latency but it may be improved upon by multi-threading and as a result, there would be no real cost on the program.
This comment was marked helpful 0 times.