Link Search Menu Expand Document

#03 ADAPTIVE CLUSTER

This server is connected to an on-grid solar system. Instead of using a battery the idea is to match energy use to energy availability: when the solar system is producing it steps up computer resources, when the solar system is not producing and grid energy is used, computer resources are reduced.

cluster

Raspberry pi Cluster

The prototype server is a cluster (bramble) of 4 Raspberry Pi 8g with a PoE hat each connected to a manageable Cisco Smart Switch 250 (CBS250-8PP-D). The switch can manage power to each PoE port, activating or closing down individual parts

Switch automation and monitoring

The automation for powering on and off of individual devices based on solar energy is composed of several layers:

  1. Solar energy monitoring: The first step is to access solar production data. This will depend on the solar setup
    • Directly from the solar inverter (production brutto) if connected, through the API that the manufacturer provides.
    • From the local energy meter using a HAN port (production netto) if this is available.
    • With a cable clamp and a microcontroller if no other option is available.
  2. Switch automation: The second step is to have a switch that can be automated based on the solar production. To be able to do this the switch needs to be a)PoE b) managable b) provide a CLI interface. Once this is availble the switch can be programmed for automatic turning on and off based on the solar data. See instructions for how this was done for a Cisco Smart Switch 250.
  3. Switch energy monitoring: Energy monitoring if using PoE can be done directly from the switch, see instructions for how this can be done in a Cisco Smart Switch 250.

Adaptive strategies

The adaptive cluster could be used in a number of different ways:

Kubernetes

A straightforward way to use a cluster is to run kubernetes containers, with the resources available for the application increasing or decreasing automatically as nodes are turned on or off.

Batching

Another option to use an adaptive cluster is to have a master node running applications nd batching heavier jobs to be performed in the variable nodes as they become available. Examples of jobs to be batched include machine learning algorithms or training, media processing, model calculations… This option need also control for not turning off nodes while hot.

The current cluster in which this website is hosted can be accessed here.