simulacra

My personal decentralized grid computing network

This repository contains the declarative blueprints for my self-organising network of heterogeneous machines and the tasks they perform. The network consists of laptops, desktops, (cloud) servers, and mobile phones.

Concepts

These are some of the ideas that I want to try to implement with this project eventually. Previously I've created raptiformica and jobrunner to help facilitate an environment in which these ideas become easier to implement in a decentralized environment with ephemeral and fast moving machines like mobile devices. This project builds on top of those tools.

Dynamic resource allocation

If a machine is online and can be reached by any of the other machines in the network, the resources should be available to the cluster. If not enough resources are available for the tasks in the cluster, more machines should be made available (autoscaling), either through Wake-on-LAN of known hardware or by booting cloud servers. Conversely, when there is not enough work to justify the amount of running machines, the grid should shut down superfluous machines to save electricity.

Resource based task scheduling

Depending on the current size and capacity of the grid, some tasks or services might become feasible to run or come online. For example, it does not make sense to run a Jenkins for testing if the grid is surviving on a cluster of mobile phones. But if in the cluster a hypervisor exists, that might be used to perform such tasks. In that case it would make sense to schedule that service by means of some automatic mechanism.

Zone based task scheduling

While the network is completely decentralized and all nodes can reach all nodes as long as any node can reach any node (through the overlay network), it would still make sense to run services in duplicate in different zones. If a couple of servers are in one zone, and some other services are in some other zone and the connection between these two networks would be severed, a netsplit would occur. In that case it would be better if services that require that level of redundancy would be preventively scheduled with one in each zone by some heuristic. Otherwise the chance is higher that two instances of a service are scheduled in the same half of a split network.

Proxying peripheral resources and services

There are parts of my network on which I do not want or can't run raptiformica natively. These are things like web services on embedded machines like IOT devices or machines not running Linux. In those cases the devices that can access those services through their local network should make those available to the rest of the machines by proxying it through the overlay network.

Usage

Clone the project:

git clone https://github.com/vdloo/simulacra
cd simulacra

Create the venv and install the dependencies

. activate_venv

Install system-wide

make install

And then create the virtualenv and install the dependencies as described here.