3 ms·
There's a few things going on. Part 1, don't do things that you should know won't survive a year of growth when you could do something that you think will at a
by toast0 4y ago
There's a few things going on.
Part 1, don't do things that you should know won't survive a year of growth when you could do something that you think will at around the same cost. Sometimes people interpret "don't optimize prematurely" to somehow mean don't do things right the first time; if you can do it right the first time, you'll save some headache down the road.
Part 2, if you build a complex product, it's hard to predict which parts of the product will see the most use, even if you can predict your overall growth rate (which is hard too). The longer you wait to make your product scale, the more information you have on what is actually used, and what you actually need to scale.
Part 3, if you make some reasonable choices, with today's server hardware, you can do an awful lot of work on a single machine. Current generation Epyc is 96 cores per socket, and AMD says 6TB per socket as well, although 3TB per socket might be more realistically achievable. You can do an awful lot of computing with a two-socket, 192 core, 6TB machine, it will be expensive, but writing a check is a lot easier than sharding a relational database. Get a database server, maybe a big one, and horizontally scale the frontend is a sensible basic architecture that gets you pretty far these days.