Joined: 21 Sep 14
AWS represents a great value for running BOINC but there seems to be very little support out there to help users in this regards.
As of the time I am writing this message; Windows based, GPU spot instances are costing me around $0.10/hr and Amazon Linux based, GPU spot instances run around $0.07/hr (Both based on the g2.2xlarge model in us-east AWS region).
The GPU based spot instances have a NVIDIA GRID GPU that crunches through work at an amazing pace. Leveraging AWS you have no expensive hardware to buy, no jacked up electricity bill, and no super heated rooms from your blazing hot GPU's & CPU's!
Sounds like a win-win yes! - Well not quite... Setting up a machine isn't as cookie-cutter as it needs to be to encourage mass use. I am amazed there are no pre-configured AMI's that have OS, GPU drivers and the BOINC Client pre-installed.
Creating a $0.10/hr windows based machine is relatively straight forward but still requires the GPU drivers, VNCServer & BOINC to be downloaded and installed manually.
The cheaper $0.07/hr Linux based solution is extremely difficult to get running. Amazon Linux instances support the GPU driver very well but getting BOINC installed has stumped me (legacy library requirements)! UBUNTU instances support BOINC very well, but getting the GPU drivers correctly installed & recognized by BOINC is a nightmare!
Is anyone proficient with AWS who is up to the challenge to create some Community based AMI's for both Windows and Linux?? I can provide assistance on the Windows side including documenting best practices for BOINC settings.
I have a running AMI template but i am not proficient enough with AWS to make it vanilla enough to work for anyone.
On the Linux side I have some existing scripts that work well for creating Ubuntu instances without the need for custom AMI (but they only run BOINC CPU tasks). I have not been able to get the GPU side operational with either Amazon Linux (Fedora) or Ubuntu - but can provide information on a plethora of combinations that don't work!
Anyone interested in working together to create a how-to thread and ultimately a set of Community AMI's?
Copyright © 2020 University of California. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.