In program 1 (Mandelbrot) of Assignment 1, the cost of each piece of work (each row) was known ahead of time. Because of this, a static assignment was appropriate, and we could predict the performance of a static assignment based on which rows of the image were assigned to each processor.
(Note that my Mandelbrot example doesn't match the example illustrated on this slide, since not all rows in the Mandelbrot image had the same cost to compute.)
This comment was marked helpful 0 times.
kayvonf
@mschervi: You are on the right track, but are you sure you knew the cost of a row ahead of time in Assignment 1? Weren't some rows significantly more expensive than others, and it took rendering the picture to figure that out? I suppose that if you ran the program once, you'd know the costs, and then you could use that information if you ran the program again on the same input. (But that's not particularly helpful, since you're using the answer to compute the answer!)
A static assignment was appropriate in Assignment 1, program 1, but I don't believe you got by with static assignment because you knew the cost of each row up front. Care to clarify? (Hint: consider example 3 on the next slide.)
This comment was marked helpful 0 times.
evan
It wasn't that we knew the cost of each piece of work, but that we knew the distribution of work. It was necessary to assign work such that the jobs (rows) given to each core weren't biased towards a particular part of the total work distribution (the center).
It seems like a very powerful strategy, as it's probably very realistic to know something about how work is distributed over a set of jobs.
In program 1 (Mandelbrot) of Assignment 1, the cost of each piece of work (each row) was known ahead of time. Because of this, a static assignment was appropriate, and we could predict the performance of a static assignment based on which rows of the image were assigned to each processor.
(Note that my Mandelbrot example doesn't match the example illustrated on this slide, since not all rows in the Mandelbrot image had the same cost to compute.)
This comment was marked helpful 0 times.
@mschervi: You are on the right track, but are you sure you knew the cost of a row ahead of time in Assignment 1? Weren't some rows significantly more expensive than others, and it took rendering the picture to figure that out? I suppose that if you ran the program once, you'd know the costs, and then you could use that information if you ran the program again on the same input. (But that's not particularly helpful, since you're using the answer to compute the answer!)
A static assignment was appropriate in Assignment 1, program 1, but I don't believe you got by with static assignment because you knew the cost of each row up front. Care to clarify? (Hint: consider example 3 on the next slide.)
This comment was marked helpful 0 times.
It wasn't that we knew the cost of each piece of work, but that we knew the distribution of work. It was necessary to assign work such that the jobs (rows) given to each core weren't biased towards a particular part of the total work distribution (the center).
It seems like a very powerful strategy, as it's probably very realistic to know something about how work is distributed over a set of jobs.
This comment was marked helpful 2 times.