Sciencemadness Discussion Board
Not logged in [Login ]
Go To Bottom

Printable Version  
Author: Subject: Decentralized Computing Power for Accessible Computational Resources
Triflic Acid
Hazard to Others
***




Posts: 486
Registered: 27-9-2020
Member Is Offline

Mood: Slowly Oxidizing into Oblivion

[*] posted on 21-12-2022 at 18:31
Decentralized Computing Power for Accessible Computational Resources


Hello again,

It's been a while since my last post, still alive; just busy. I now work in a university synthesis lab and have been doing some computational work recently to model some compounds I synthesized. I was thinking about the scope of computational work these days, and how nearly every chemist regardless of specialty will eventually have to run a DFT/geometry optimization, etc. At our uni, we have access to a node-based supercomputer to run computations, but many smaller universities/people like us doing stuff from home can't access these resources.

I've personally run orca computations on my laptop, which ended with it overheating and crashing when I tried to DFT an 8-unit polymer for work since the VPN to connect to the supercomputer was down. Running it on the supercomputer took a few minutes, so that ended my at-home computational chemistry. But I was wondering if this sort of idea would be possible:

People online can sign up to join the computing service, getting paid $1-$2 per hour of CPU usage.

Users online can buy hours/part thereof of computational time, pricing enough to break even based on what the computation cost.

A central computing node takes the computation from the users, breaks it down, and distributes it evenly in packets to all of the active computers.

The center hub then stores the results, receives cash from the customer, then distributes the cash to the computers at the usage rate.

So, horrible idea or genius? I know that systems like AWS exist, but they are prohibitively expensive for someone entering if you want to run actual computations. Stuff like DFTs will cost you a $2500 monthly subscription from what I read. If this really turns out to be something the community is interested in, I'll write some code and try to get it to work.

Thoughts the suggestions appreciated

-Triflic




There wasn't a fire, we just had an uncontrolled rapid oxidation event at the power plant.
View user's profile View All Posts By User
Sulaiman
International Hazard
*****




Posts: 3508
Registered: 8-2-2015
Location: 3rd rock from the sun
Member Is Offline


[*] posted on 21-12-2022 at 21:22


Try a better laptop OR don't overclock yours.
No matter how computationally intensive an application is,
your laptop should not overheat and/or lock up.
If it takes a couple of days to complete it's task - plan ahead.




CAUTION : Hobby Chemist, not Professional or even Amateur
View user's profile View All Posts By User
Triflic Acid
Hazard to Others
***




Posts: 486
Registered: 27-9-2020
Member Is Offline

Mood: Slowly Oxidizing into Oblivion

[*] posted on 21-12-2022 at 23:06


I managed to get mine to work after that, but it was still pretty long calculation. I'll pretty much never use this again, but also there is a upper limit as to what a single PC can do reasonably. I only managed to get the 4 unit to converge, and after 72h the 8 unit hadn't converged. I can't afford to not use my laptop(was too laggy to use for the duration of the jobs) for that long, so I gave up and waited for the VPN to be fixed. The supercomputer did it in 8.56sec



There wasn't a fire, we just had an uncontrolled rapid oxidation event at the power plant.
View user's profile View All Posts By User

  Go To Top