The title is a bit generalized statement of what I'm trying to do. Let me explain...

I'm working on a project for an Amazon Web Services certificate program. In this project, I am working on a proof-of-concept to take advantage of AWS "Spot Pricing". Essentially, during periods of low usage, AWS makes EC2 instances (a cloud optimized VM) available for 10 to 25 percent of normal rates.

My project will periodically poll these spot prices and when they reach a preset point, the project will create a set number of EC2 instances to run s@h clients for a period of time. When the spot prices rise to a set point or above, the project will stop the instances. The instances will created from a template with a boinc-client already installed. An instance's unique boinc information, current work unit, and the work unit's state state information will be copied to an instance when it created and copied off of the instance as part of its tear-down process.

What I'm looking for is a reference of what information needs to be copied onto an instance so that it will be able to pick up and start working on a work-unit where a previous instance had left off. I will have set these to not cache work units and only have the work unit being processed be present on an instance.

Currently, I am thinking I will create 5 machine accounts and have 5 copies of unique information set up beforehand. Is it possible (or advisable) to have each instance work with the same machine account?

Thanks,

Bruce

---

Seti tasks can be calculated on any system, they aren't hardware dependent. Science applications that do run the tasks are hardware dependent.
Task downloaded by one BOINC must be uploaded and reported by that same BOINC. They cannot be reported by any other BOINC, because tasks are registered against the hostID.

So then, to be able to run tasks downloaded by one system on another system, you should move the whole data directory, including all science applications, files and directories, to the other system & reinstall BOINC on that. The reinstalling makes sure that permissions on the data directory are set correctly.

A couple of cruxes:
- hardware needs to be the same, or same-ish: Intel CPU vs Intel CPU, AMD CPU vs AMD CPU, Intel/AMD/Nvidia GPU vs Intel/AMD/Nvidia GPU.
- operating system must be of the same class: Windows vs Windows/Linux vs Linux/Mac OS vs Mac OS/Android vs Android.

While it may be possible to run tasks downloaded for an Intel CPU on an AMD CPU, any GPU or any coprocessor, it cannot be reported by such a system because the host consists of a specific CPU vendor and/or GPU vendor registered under that hostID.

---

Thanks Jord. That information helps.

I'll be spinning a VM based on a template. The template will already have BOINC installed and I'll be spinning up VM's of the same class. This will keep me from needing to install BOINC each time a VM is started and the VM will appear to be identical hardware-wise.

The template is configured to not start BOINC at startup. This will allow the data directory to be copied into place and then I can start BOINC. The VM will get two minutes warning of a shutdown. This will give the VM enough time to copy data directory to a network share before it is terminated.

I still have some AWS pieces to work out but I should have enough BOINC info for that part to work.

Bruce

---