Carbon footprint of Algorand

General
1

Now, that Elon Musk has made his 1.5 billion dollar investment to bitcoin, his critics say that it is “an environmentally dangerous move”.

According to a study published by science journal Joule, The Carbon Footprint of Bitcoin was about 23 megatons of CO2 in 2018. “Those crypto-emissions are equivalent to the CO2 spewed by well over 4 million gas-powered cars each year”, writes Russ Mitchell in a Los Angeles Times article.

I googled “Algorand carbon footprint”, but I didn’t find any data.
I wonder, what are the relevant data for Algorand? And what about future, with 46000 TPS?

2

@Maugli,

As I’m sure you already aware, the Algorand network consumes far less energy ( by many order of magnitudes ) compared to Bitcoin ( or Ethereum ). I wish I had the exact numbers, but I do not.

In fact, it would be really hard to get exact numbers :

  • Many of the nodes aren’t managed / owned by Algorand Inc and/or the Algorand Foundation.
    ( for example, I have no idea what your machine specification are )
  • The is a huge difference in the power consumption of a gaming PC ( ~ 400W ) vs. a raspberry PI ( 3W ).
  • Some of the machines used as relays might be hosted by aws/azure/gcp. For these, you’re getting a “portion of the machine”. I don’t know how to gauge the Wattage on these.
  • For development purposes, we continuously deploying dummy test networks. These networks are ephemeral, but could consume lots of power while running. I don’t have any metrics for these.
  • Last, many of the machines that we use are far over spec’ed ( just to ensure we have that capacity if we end up needing it ). That’s a wasted energy that we knowingly willing to spend as long as we’re making changes to the product.

Having said all the above, I will try to look if we have any concrete numbers that can be used here. For simplicity, I would rather using the current network metrics, as it’s factual. ( i.e. and for the same reason, I’d like to avoid discussing about Eth2 - it’s only speculations for now )

You can start looking here https://metrics.algorand.org for some metrics. On thing you might to grab from there is the peak TPS, which currently is 1,161. That would dynamically change as the network throughput increases.

3

Hi @Maugli,

I would like to add some considerations following what @tsachi mentioned.

Since the estimation of the carbon foot print strictly depends on power generation scenarios and the degree of renewables sources, I would rather address another critical metrics: the specific finalized transaction energy, that is the amount of energy spent by the whole blockchain infrastructure to finalize a single transaction and append it to the common distributed ledger.

I want to emphasize that referring to finalized transaction in the definition of such a metric is not just a mere subtlety of terminology. Proof of Work consensus protocols, in fact, are prone to “soft forks”. When this unfortunately happens these blockchains split in branches representing temporary inconsistent multiple sources of truth within the same ecosystem. While the miners keep working on that chains, providing their respective hash rate and wasting electrical power, just one single chain is going to survive over time, according to the principle of “the longest chain”: the chain on which the greatest amount of computational work has been done will keep growing, leading all the others to an impasse. The probability of soft forks is not neglectable in PoW protocols, therefore transactions in PoW blockchain are commonly considered “final” only when a 6 blocks depth is reached. The whole amount of energy spent in validating transactions belonging to “orphan chains” is completely useless, moreover all the energy wasted on these transactions must be spent again, until they end up being appended to the longest chain. This makes PoW even more inefficient, leading some of the biggest PoW blockchains to have energy footprints that rival small countries.

I propose here an estimation of such metric that relies on very rough assumptions and its purpose, rather than a rigorous analytical calculation, has to be intended as a reasonable baseline to support discussions on Algorand’s energetic efficiency with reasonable data evaluated within a coherent hypothesis frame.

Hypothesis

  1. Relay Nodes and Participation Nodes power consumptions are comparable;
  2. PPoS keeps same Participation Node hardware requirements regardless the number of nodes participating in the consensus;
  3. PPoS keeps same Finalized Transactions per Second (TPS) regardless the number of Participation Nodes in Algorand Network;
  4. Participation Node running process is accountable for the whole power consumption of the hosting hardware, regardless of whether PPoS TPS are used at full capacity;
  5. Internet energy consumption has not differential impact between different blockchains architectures.
  6. Node Synchronization and Storage power consumption is not taken into account;

Since Algorand has no way to determine which kind of hardware hosts each Participation Node in the network, it is reasonable to assume that rational people would tend to choose the most efficient hardware capable of running the node process while ensuring full capacity TPS (e.g. a Raspberry Pi 4).

Participation Node Minimum Hardware Power
Pn = 7,2 [W] per node (e.g. Raspberry Pi 4 at full capacity)

PPoS Finalized Transactions per Second
TPS = 1000 [txn/s]

Finalized Transaction Energy per Participation Node (Etn)
Gross amount of energy spent by a Participation Node for each finalized transaction

eqn_1

Dimensional Analysis

eqn_2
eqn_2916×227 7.38 KB

Conclusion
Considering that:

eqn_3

We can conclude that, on average, for “low-end” Participation Nodes (e.g. Raspberry Pi 4):

eqn_4
eqn_4864×304 8.27 KB

Algorand Finalized Transaction Energy
Under the hypothesis, the amount of energy spent by the whole Algorand Network for each finalized transaction grows linearly with PPoS degree of decentralization:

Et = Etn x (Participation Nodes) = 2 [𝜇Wh] x (Participation Nodes)

Considering different scenarios for the Algorand Network “low-end” Participation Nodes:

Algorand, 1k nodes
Et = 0,000002 [kWh/txn]

Algorand, 100k nodes
Et = 0,0002 [kWh/txn]

Algorand, 10M nodes
Et = 0,02 [kWh/txn]

Some data on this metric for Bitcoin:

Bitcoin
Et = 650 [kWh/txn], not all final

The following results must be intended only as gross estimation of the order of magnitude for the amount of energy spent by Algorand for each finalized transaction.

4 Likes
4

Hi, Cusma,

Thanks for the detailed answer.
Of course, instead of RPI you can count with Intel Xeon processors, etc, with roughly a hundred times as much power consumption. Still, it would give

Algorand, 1k nodes

Et = 0,0002 [kWh/txn]

which is VERY IMPRESSIVE.
1 Like