http://www.andrewheiss.com/blog/2018/07/30/disposable-supercomputer-future/

[andrewheiss]

Create a cheap, disposable supercomputer with R, DigitalOcean, and future | Andrew Heiss
http://www.andrewheiss.com/blog/2018/07/30/disposable-supercomputer-future/

[andrewheiss]

Originally posted by César Arquero Cabral on 2019-04-30 11:28:37

Hi, I'm trying to implement exactly this but I'm stack in the
"ssh_private_key_file <- "/Users/andrew/.ssh/id_rsa" part.
This is the only file that I have within that dir:
"ssh_private_key_file <- "/home/cesarkero/.ssh/known_hosts"
When I execute the makeclusterpsock functions it appear this message:
WARNING: UNPROTECTED PRIVATE KEY FILE!
I've tryed to redirect the ssh_private_key_file to both private and
public ppk made with puttygen...but nothing works.
Do you have any advise ?

[andrewheiss]

Originally posted by César Arquero Cabral on 2019-04-30 11:31:38

Here's is what I get:
https://www.evernote.com/l/...

[andrewheiss]

Originally posted on 2019-04-30 15:24:08

You need to adjust the permissions for ~/.ssh:
[https://www.howtogeek.com/1...](https://www.howtogeek.com/168119/fixing-warning-unprotected-private-key-file-on-linux/ "https://www.howtogeek.com/168119/fixing-warning-unprotected-private-key-file-on-linux/")

[andrewheiss]

Originally posted by César Arquero Cabral on 2019-05-02 18:13:31

I have finally resolve it and now i can conect to those machines. How
could i do now a process that uses a huge csv (raster) data within the
main machine? Does the data need to be downloaded into the
workermachines ?

[andrewheiss]

Originally posted on 2019-05-02 18:14:58

Yeah, the data has to be present on each of the machines

[andrewheiss]

Originally posted by Paul John on 2019-11-29 17:30:43

How can i implement this if i have owned DigitalOcean Hosting from
Cloudways platform. My main application is hosted there but too much
space available to try this out

[Tadge-Analytics]

Is it possible to use the cores across the remote cluster as well as the cores of the local machine (all at the same time?) -great + very hepful blogpost btw 😁

[WihanZA]

Hi Andrew. Thanks for the very useful post. I was wondering: If each docklet is running Ubuntu (a POSIX system), would it not be more preferable to use a makeForkCluster()? Also: Could you please shed some light on your decision to use plan(cluster, workers = cl) as opposed to some other plan, such as those using tweak()? Please excuse my ignorance if I'm missing something obvious. Thanks again for the good work!

[andrewheiss]

Mostly because I wrote this in 2018 in the early days of the {future} package, before other functions like tweak() existed :)