This was fascinating. If any Go compiler experts happen to read this and know why `Entry{}` was copied on the stack before being copied into the buffer, I think that would be cool to read and tie off this article nicely
Well, the author's investigation was involved, although it seems like a labour of love, and I enjoyed reading it.
If benchmarking 28 vs 32 didn't yield an obvious result, lazier old me, if I'd even been benchmarking at all, might have tried more data points, hoping to reveal a trend line that points to somewhere interesting e.g. 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32.
Why would that be obvious? Decentralized means losing economies of scale - you can't buy 10k drives at a time from the manufacturer, you run on random hardware builds, you pay to run machines on lower-efficiency PSUs, and you have to compensate for worse uptime per node. Where is it winning?
Decentralized storage taps into latent/underutilized capacity from datacenters and individuals across the world. For example, if you have 10TBs of free capacity that wouldn't otherwise be used, and running a node is pure margin. Most private datacenters are at about 40% capacity.
Another example –> the total cost for me is more than 10x cheaper than AWS - its 7$ for TB/Bandwidth vs $90 on AWS. I can also get 1.8gbps - in my region, which is even faster than AWS, and it doesn't rely on ACL databases (it uses client-side hmacs for file authorization). So, for my use case (static content storage and low-volume CDN usage) decentralized cloud wins. (it's 10x cheaper, faster, and more secure for my needs).
