This one seems to have slipped by me, but almost a week ago Backblaze posted on their blog a report they had derived about Chia farming in their datacenter. It is a really good read, and if you are interested in running storage at scale for Chia farming, and the challenges therein, its a really good read.
They did a lot of investigation into the technology, including the read patterns of a proof lookup and what you are in for if you decide to use cloud storage for Chia plots. It is quite an extensive guide, and matches up with what I have been saying about the challenges of running gigantic amounts of storage reliably and available for quick lookups.
The Backblaze team has gone so far as to dig through the chia-blockchain codebase and find a pull request that would improve parallel request access and made some changes to their Fuse filesystem to cache the specific headers in order to get the plot checks down into the required window to win plots. Its obvious they put some real effort into this, and this was not just a side project for a bored and greedy sysadmin. It really looks like they got it working, with a combination of code changes and a deep understanding of enterprise storage.
The fact that they had to make these changes in the first place supports my hypothesis that there are large farming operations out there running so inefficiently that they are consistently losing rewards. If it took Backblaze some skull sweat to figure this out I cannot imagine every operation out there has managed the same.
I’m not sure about the economics of running plots on cloud storage you pay for monthly, but it seems like an interesting experiment. If someone is reading this and is running a large scale operation off public cloud storage I would love to sit down and talk to you. Reach out to me directly or leave a comment here.