Some data might be more computation-intensive as compared to other. Calculating square root using Newton's Method in Program 4 (Assignment 1) is an example of this. Keeping this in mind can help in avoiding work imbalance.
sadkins
In the mandelbrot image from Program 1 of the homework, partitioning the work into equally sized sections of the image(data) is not an even distribution of work because the white pixels, which are concentrated in the center of the image, require more work to compute. Therefore in this case it does not make sense to break up work by data
lfragago
I think in the case of homework it DOES make sense to break up the work by data. The important thing here is HOW to assign that work. Depending on the work assignment policy we use we will get better or worst performance.
Metalbird
I think the point gets back to even though we think of it as how are we decomposing the data, the motivation behind the decomposition strategy is to distribute the work evenly. Even if we have a case where some data requires more computations than other sets of data, we have to make sure that our decomposition policy keeps that in mind and divides the work accordingly. So while we may consider how to decompose the data, we are really decomposing the work into separate tasks.
rc0303
Sometimes, computation is done with multiple data sets that have different meanings. For example, in the renderer for Assignment 2, computation can be split across two axes: pixels and circles.
Some data might be more computation-intensive as compared to other. Calculating square root using Newton's Method in Program 4 (Assignment 1) is an example of this. Keeping this in mind can help in avoiding work imbalance.
In the mandelbrot image from Program 1 of the homework, partitioning the work into equally sized sections of the image(data) is not an even distribution of work because the white pixels, which are concentrated in the center of the image, require more work to compute. Therefore in this case it does not make sense to break up work by data
I think in the case of homework it DOES make sense to break up the work by data. The important thing here is HOW to assign that work. Depending on the work assignment policy we use we will get better or worst performance.
I think the point gets back to even though we think of it as how are we decomposing the data, the motivation behind the decomposition strategy is to distribute the work evenly. Even if we have a case where some data requires more computations than other sets of data, we have to make sure that our decomposition policy keeps that in mind and divides the work accordingly. So while we may consider how to decompose the data, we are really decomposing the work into separate tasks.
Sometimes, computation is done with multiple data sets that have different meanings. For example, in the renderer for Assignment 2, computation can be split across two axes: pixels and circles.