Friday, May 15, 2020

Economics Of Decentralized Storage

Almost two years ago, in The Four Most Expensive Words in the English Language , I wrote skeptically about the economics of decentralized storage networks. I followed up two months later with The Triumph Of Greed Over Arithmetic. Now, Got a few spare terabytes of storage sitting around unused? Tardigrade can turn that into crypto-bucks is Thomas Claiburn's report on initial experience with Tardigrade, the "decentralized" storage network from Storj Labs. Below the fold, some more skepticism.

First, lets look at pricing. Claiburn writes:
Last November, when pricing was announced, storing 1TB for a month cost $10, with egress bandwidth charged at $45 per TB. As a point of comparison, Amazon S3 charges $23 per TB for a month and $90 per TB, subject to significant variation. The Tardigrade network currently has about 19PB of capacity, 12PB of which are used, from about 6,500 platform participants.
But comparing decentralized storage to S3 is comparing apples and oranges. S3 is not simply a storage system. It is a part of a complex, integrated computation ecosystem, and is priced against the value that the ecosystem delivers.

The correct comparison is with storage-only services such as Backblaze and Wasabi. With that comparison, my skepticism looks justified. Wasabi charges $5.99/TB/month with $0/TB egress charge.  Backblaze charges $5/TB/month with $10/TB egress.

So, if you never access the data, Tardigrade is twice as expensive as the centralized competition. If you access 50% of the data each month, it costs $32.50/TB against Wasabi's $5.99, so more than 5 times as expensive. What exactly is the value Tardigrade adds to justify the extra cost to store data? Simply "decentralization"?

But, like all these cryptocurrency-based systems, Tardigrade's "decentralization" is more a marketing term than a practical reality. The money isn't decentralized, because customers pay Storj, who then pays a little of that money to the storage node operators (SNOs):
According to a company spokesperson, SNOs earn $1.50 per TB per month for static storage and $20 per TB for egress bandwidth, with the caveat that actual compensation varies based on network utilization, internet connection, and geographical location.
Storj keeps 56% of the bandwidth revenue. Their erasure coding scheme appears to have a 2.66 replication factor, so for every $10 TB a customer stores for a month, the network needs to pay SNOs at least $4. It isn't clear whether the $1.50 is for a TB of capacity, or a TB of stored data:
  • If it is stored data, Storj keeps 60% of the storage revenue but also controls how much of your capacity actually earns for you.
  • If it is capacity, Storj only keeps 36% of the storage revenue.
Either makes even Uber's advertised (but misleading) 25% look generous. Note that the $1.50 isn't even actual money:
For sustaining their nodes at 99.9 per cent availability, SNOs receive a token payment in STORJ tokens.
STORJ tokens are worth about $0.10 at present, though they must first be exchanged for another cryptocurrency like Bitcoin or Ethereum before they can be turned into a fiat currency like US dollars.
STORJ, with a "market cap" of $14.4M, is a rather small cryptocurrency, thus its value is highly volatile and easily manipulated. The SNOs bear these risks, and have to pay fees to trade into a larger cryptocurrency and more to trade into fiat. Even the company admits that, if you have to pay for the media, power, etc. the economics of running a node don't work:
In other words, Tardigrade isn't a storage-rental-to-riches scheme. Gleeson said the company does not recommend that people buy storage just to become a SNO. Rather, the network provides a potential way to defray the cost of existing storage hardware.
The company's ideal SNO is a data center that can't manage its procurement properly:
"There are a lot of partners with data centers that are underutilized and are seeing they can use that capacity to add to the profitability of their businesses," said Gleeson.
And, lastly, there are actually two kinds of nodes in the network, storage nodes and Satellites, which manage access to them:
At production launch, the Storj network will allow anyone to operate a Satellite in addition to the Tardigrade Satellites operated by Storj Labs. Being a Satellite operator is a big commitment with very stringent uptime requirements, much more so than a storage node operator. If a storage node goes offline temporarily, data can still be recovered from other nodes. If a Satellite goes offline temporarily, the data stored by that Satellite could become unavailable. Satellites are also responsible for managing payments to storage nodes storing data for the Satellite as well. 
Note that the Satellite through which your data was stored is a single point of failure.

Lastly, based on both theory and practice, one would expect that the vast majority of Satellites would be "operated by Storj Labs", so there would not really be "decentralization". This is, in fact, almost a legal requirement:
"If someone publicly shares illegal or copyrighted content and publicly shares it, if that is brought to our attention through a valid court order from law enforcement, we will comply with the laws and jurisdictions to remove that illegal content from the network that is connected to our Satellites."

"This hasn't happened yet, but we have a system in place to deal with these types of requests," he added. "However, we're decentralized and open source, so anyone can run a Satellite. We're working on a framework to help others who want to operate Satellites navigate this, as laws and regulations are very different across different regions and countries."
The history of ICOs shows the folly of assuming that "decentralization" is a magic shield protecting you from legal assault.

1 comment:

  1. I am currently using Wasabi to digitally store the Lincoln Archives Digital Project records ( Have been tremendously pleased with Wasabi. Since I have over 100 hard drives, Wasabi has been speedy to upload files, with no penalty for downloading.