Open anders34 opened 8 years ago
My discussion / thoughts on the chat board:
At this point in our HPSC education we are beginning to realize that there are many many factors that go into how well our code runs on a given machine. My code runs differently on my laptop, which only I use and has a different processor, than the 300+core SMC machine which probably uses a different processor with different layers of task scheduling. You can spend days reading about how cgroups work. And about the specific architecture of a CPU And about how Argonne National Lab's OpenMP library (which is used on SMC) differs from the OpenMP implementation that you might be using on your personal machine. It's all really interesting to think about!
My point being, think about all the things we talked about in class and in the Issues regarding parallel code on massive machines (and maybe even do some research on your own on the internet!) and come up with a somewhat logical explanation. The number of possible directions you can go with the hardware and software aspect is staggering, to say the least.
One question we're supposed to answer is, What can be attributed to the nthreads = 8 behavior when the chunk_size is small? I have two questions regarding this: 1.I have odd behavior for all of nthreads=8, not just when chunk_size is small, but for every chunk size. Is it not supposed to be like that? (I'm not sure if actually saying what the odd behavior is is too much information)