/tech/ - Technology★

It depends on what you want to parallelize. A common problem is calculating a matrix for which you later want to get the eigenvalues.

Usually the matrix elements are independent of one and another, so you can subdivide your matrix and have those segments calculated on different cores/servers. So you'd need a master program that assigns these segments to slave programs and then aggregates the results.

Of course there are libraries that abstract this process to varying degrees, but one language that was actually designed with this in mind would be Julia:

https://docs.julialang.org/en/stable/manual/parallel-computing

Install Gentoo on the servers

OP do this >>828542

>>828537

Whois of (((http://zguide.zeromq.org/)))

>Name:

>Doron (((Somech)))

>Street:

>3 Liberman St.

>Apartment 10

>City:

>(((Tel Aviv)))

>Postal Code:

>6424303

>Country:

>IL

>Phone:

>+972.543044117

>Email:

>email@gmail.com

You couldn't make this shit up if you tried.

>How the fuck do people write programs that run across multiple machines at once?

Network protocols.

>>828502 (OP)

Networking. The single machine code is the same as everything else, then you have something that handles the rest of it. MPI used to be popular for that but I always thought it was kinda shit. The advantage of using it was industry supported it, so you could get various hardware accelerated networking devices for MPI.

>>828545

a) what's your point?

b) it's called Whois protection. Look it up.

>>828542

Plan 9 is better choice tbh.

>>828524

>slave programs

<TRIGGERED

Super problematic terminology in 2017.

https://code.djangoproject.com/ticket/22667

>>828502 (OP)

Look for books on distributed computing. Essentially though, the concept is to atomize the bigger task into units which can be farmed out your machines, and then aggregated into a solution.

>>828502 (OP)

Google PVM (parallel virtual machine) and MPI (message passing interface)

You can also use a very coarse-grained approach using a server/client model where all computers share a common server. There is a shared directory on the server with two subdirectories; an in and out-box. The client pulls data from the out box and places the processed returns in the in box. The server handles shuffling files from those directories to divy up the workload. I've personally done this twice in practice and would not recommend it.

Gnu parallel can run several instances of a program on a cluster.

>>828542

install guix

>>828770

>>828778

>>828761

>>828703

>>828606

>>828537

>>828524

Thank you for all the helpful answers, this is a great start for some googling. Do any anons have any other info or stories they'd like to share about cluster computing?

>>828545

>>828630

kek

>>828826

If you just want to fuck around, it's pretty easy (or, at least it was back in 2014) to get luxrender running on multiple machines. I had eight seperate computers of *very* different architectures rendering a single image, three of which were using OpenCL acceleration on top of the fact that they were clustered. It was a very fun little experiment.

>>828913

Well fuck.

Just checked and LR is pretty much dead. The website is still there, but there's only one or two people who've touched the forums in years and the download links aren't working. Shame.

>>828826

>Do any anons have any other info or stories they'd like to share about cluster computing

I used to work in supercomputing (which was often just cluster computing) over at LANL and so did my girlfriend. This was back when tech was 1% female and before cam sites so you'd sometimes run into a legitimately competent woman, maybe '95-'96? She'd not come around one night and had terrible excuses as to why and I figured she was probably being a cheating whore, but I learned later that her group of friends had planned on watching a movie after work and decided to borrow a DVD player from the lab. They were somewhat new and expensive at the time (would be several thousand today) so none of them had one. The lab had a big security team known as proforce (great name) who were looking for chinks smuggling out anything that might store data. So that had triggered a full security incident complete with cliche black vans driving them off the road, attack dogs, snipers, helicopters, and a day of interviews. She was a cheating whore, though. Don't trust women in tech.

I hope I helped you learn a bit about cluster computing!

>>828703

>that django ticket

fuck my life i use django. dotr when

It's all essentially just simultaneous multi-threading.

>>829273

>simultaneous multi-threading

No, that's a totally different thing.

>>829309

Ehh. You could, but you can just test it as a pretend thing with multiple processes per host each setting affinity to one full core, potentially sharing with more processes just to test. imo you're gaining nothing but trouble from trying to use VMs.

>>829308

???>>829388

>>>829309

>Ehh. You could, but you can just test it as a pretend thing with multiple processes per host each setting affinity to one full core, potentially sharing with more processes just to test. imo you're gaining nothing but trouble from trying to use VMs.