Listening to the first episode of a Tech Won't Save Us podcast subseries, called Data Vampires;

https://techwontsave.us/episode/241_data_vampires_going_hyperscale_episode_1

Once again it occurs to me that hyperscaling of datacentres is exactly the wrong approach to hosting online services. Not just from an economic justice and environmental POV, but even from a business POV.

(1/?)

#podcast #TechWontSaveUs #DataVampires #DataFarming #DataCentres #HyperScaling #DataHubs

Data Vampires: Going Hyperscale (Episode 1) - Tech Won’t Save Us

Tech Won't Save Us

Imagine a hosting service that owned two rooms full of internet-connected computers, providing the hardware for hosting online services. Let's call each of these rooms a "DataHub"

One is in the southern hemisphere, one in the north. They only operate each one about 6 months of the year, based on the temperatures in each location. But at any given time one of them is always operating.

(2/?)

DataHubs are small enough to fit inside an apartment building, and the AC that cools them moves the heat it extracts to the rest of the building. Because each DataHub only operates in the colder months, they require less energy to cool anyway.

The DataHub operator pays rent to the building owners, and charges them for the heating services they provide. Or they might own the building and rent out the apartments, inclusive of winter heating, for a flat fee.

(3/?)

So, basically, we're taking electricity that we currently convert directly to heat, and running it through computers first. Making heat that way. Instead of powering two AC systems - one for the apartment building and one for a datacentre - we're using one AC system for both.

This is an application of permaculture design principles to information systems - an aspect of what permies call "invisible structures".

(4/?)

#permaculture #design #InvisibleStructures

Some of the principles applied here;

* The problem is the solution

Rather than seeing heat created by server computers as a problem, we reframe it as a resource. Allowing us to ...

* Use and value renewable resources

Obviously we'll be looking to make sure the electricity powering the DataHubs comes from renewable generation. Ideally as local as possible.

* Slow and small solutions

Both DataHubs and the generation powering them can be small and numerous, slowly replacing the giants.

(5/5)

Coda:

One of the deeper goals of the internet re-decentralisation movement needs to be abolishing hyperscale datacentres, and moving hosting to community-controlled small scale DataHubs.

One thing that would help is anti-monopoly rules that prevent companies from owning servers *and* owning businesses that rent servers. Or at least banning them from giving their own businesses preferential pricing.

#decentralisation #DataHubs #PolicyNZ

Afterthought:

The generational "AI" boom is like the biofuel boom of the early noughties. Why? Consider this.

45% of corn cropland in the US is made into ethanol;

https://www.ers.usda.gov/topics/crops/corn-and-other-feedgrains/feedgrains-sector-at-a-glance/

This is happening because the US govt have been sudsibising over-production of corn for decades. See Michael Pollan's book The Omnivore's Dilemma, and the work of David Blume, author of Alcohol Can Be a Gas.

(1/2)

USDA ERS - Feed Grains Sector at a Glance

The major feed grains are corn, sorghum, barley, and oats. Corn is the primary feed grain in the United States, accounting for more than 90 percent of total feed grain production and use.

So biofuel boomed in 2001 because it was a new way to get rid if some of the corn mountain, and as a bonus, a way of greenwashing the industrial farming industry.

Trained #MOLE hype is booming now because there's a similar mountain of overproduced commodity, looking for a use and a bit of reputation laundering. In this case, a computation mountain.

Feeding that computational corn to Trained MOLEs is a great way to make both look useful.

(2/2)