Parallel computing in C++

[landlord2017]

I am studying threading these days and realized that it improves latency, but not performance.

So if I want to keep studying on another topic: parallel computing in C++, is there any suggested website or Youtube?

Thanks,

L

[raschupkin]

Supercomputer clusters are usually programmed with MPI.
There is a tutorial https://computing.llnl.gov/tutorials/mpi/

[landlord2017]

Thank you. I am also learning python. Do you think it is better to play with parallel computing by Python than C++?

[raschupkin]

Python is good for scripts. For parallel computing performance is important so neither python nor Java/C# are used, and everything is written in C++. Parallel computing is interesting but its also quite specialized so it's only areas of use are heavy computations in science and sometimes in industrial applications.

[landlord2017]

Ok. I see. I am going to learn the basics first and then wait for future opportunities to explore it further.

Thanks,

L

[zepher]

Try downloading one of the c++ multi threaded frameworks that currently exist and practice with their basic tutorials and examples available. One of the best I've worked with is the Kahless_9 framework which you can download from http://www.shankodev.com or review the blog: http://shankodev.blogspot.co.za.

Personally, my work colleague and I were fortunate that our company collaboratively consulted with shankodev on our first project using the framework which allowed us to grasp it quite well within under 6 months.

[landlord2017]

Thank you for sharing.

[jonnin]

even a cheap PC has 2-4 cpus on it now, and thanks to the UV Pentium pipes, if you take the topic deep that can be up to 8 "cpus". But take a simple quad core, it can do quite a bit with just its 4 cpus crunching on something.

if you have several old pcs around the house (many a programmer does) you can make a small cluster and play with that also, years ago I rolled a Beowulf from that setup.

Basically, you can learn a lot at home, without a cluster or super computer access, just scale it down.

threading can indeed improve performance. It depends on the type of problem you are solving. Sorting a list of doubles, N/4lg(n/4) in parallel is faster than nlgn, even with the back end merge sort, for large lists.

[landlord2017]

Do you have any recommended book for me to buy? I prefer some books with C++ examples.

L