3 ms·
> I think the author critically under-guesses the sizes of things (even just for storage) by a reasonably substantial amount. While true, and not to take away
by Xeoncross 4y ago
> I think the author critically under-guesses the sizes of things (even just for storage) by a reasonably substantial amount.
While true, and not to take away from the parent comment, I've noticed that the size of things is often partially the result of scaling out horizontally. Most companies I've worked at end up with a lot of duplicate records as each subsystem might want a copy or to cache a copy.
- vidarh 4y agoThis is indeed a problem, and one of the reasons to be careful about knowing you can fit things in one machine before taking that approach, because the moment you're forced to move to a distributed model it's rarely one machine to two but one machine to a dozen coupled with a major rearchitecturing effort at just the wrong moment. It's often fine to start without a fully decoupled system (net present value of the time and money needed to scale out might be far too high), but you need to know whether or not it's likely to come and what to look for so you can start preparing in time.