DHH's focus getting massive saving on storage costs is a head-scratcher. Usually, egress costs are punitive. Given LLM's are pretty good now, I modeled the break-even point of such a move from S3 (6PB) to a colo. Breakeven time 18 years(!!), even with 8 year depreciation cycle still looking at 8 years to breakeven. This is not including falling S3 costs in Amazon. Truth be told he is probably spending 10's of millions on servers running Rails (SSR), probably can cut that in half by moving to golang or java. This is peanuts in comparison.
1. Initial costs of $2.5M to provision the hardware (disks, servers, enclosures, networking equipment, redundancy, software solutions etc.)
2. Facility OPEX: $50,000/month (Power, connective, monitoring etc..)
3. Staffing and Operation tools: $10,000 / month
4. Replacement cycle of 5 years so assume $500,000/year ~ $41K/month
Are you sure? It looks like they only have 1 rack (+1 in another facility for redundancy) and seem to have 40Gbit/s connectivity.
A full rack is in the range of 1k, connectivity around 600$ per 10Gbit/s. I have no idea how much power they consume but I doubt it's 40k$+ per month for a storage workload. I would guess they are in the 10k$ range. Those are only list prices I've seen in the wild so take it with a grain of salt but 50k seem VERY high.
1 rack is a recipe for failure, you need to split, even without that consider
how many SSD's you need.
power usage for all those SSD's
inter-site connectivity (need to keep transferring data between the sites otherwise customers are going to be very surprised.
maintenance and software costs (at the colo level)
All these add up, closer to $50K than some $2K (LOL). The way you guys (below) are talking, this is not some home server that serves personal videos. It runs (some) business operations for thousands of small/medium companies.
