What books to read about parallel computing locally on multiple computers?

Question

Distributed Computing

I found 3 obsolete and traditional methods in the book QNX/UNIX Anatomy of Parallelism by Oleg Tsilyurik, Egor Goroshko, published on the flibustier librusec site of a pirated library banned in the Russian Federation (partially) and published on another site.
I read a comment on the site of a large store that you can do "child process on the CPU of the local computer along with child services." These processes and services are indicated by almost zeros when idle and perform calculations in proportion to the idle resources of THREE personal computers connected by Ethernet NETWORK and continuously, having only previously distributed the initial materials over the network. The summaries and results are sent to the personal computer on which the main process is running.
The essence of the task given to me is this, the interface can be replaced with another, more productive one in descending order of bandwidth:
- connection through the south bridge (unlimited speed and depends on the motherboard) I will consider for other purchased computers before buying them,
- 24 GB per second optical Internet interface,
- 20 GB per second - thunderbolt 2,
- 10 Gb per second - PCI Express 16 bus will not be considered by me and the slowest of those indicated.
Answer the question: "Which books on parallel and distributed computing to read?"

Answer 1

unlimited speed and dependent on motherboard

The data bus has limitations, if that.
www.open-mpi.org - perhaps the best freeware for organizing parallel computing in a cluster. A kind of de facto standard.
In general, throughput is not so important, since the profit from this approach is only achieved when the execution of the task is much longer than the message transfer time. Yes, and these messages are sent over the network, and this in itself imposes restrictions. Of course, you can write and communicate directly, via DMA, but something seems to me that it will be very expensive.
ps 10 gigabits per second slowly (or 8, I don't remember how much x16 3.0 pulls out there)? I may not remember well, but the CPU bus in modern motherboards will give approximately the same indicators (or 15). Do I understand correctly that you want to make one out of three computers? Will not work.