LocustDB
DatabaseComments
Does the project provide any guidance on how it handles concurrency across very high core counts? I am curious if it uses a shared-nothing architecture internally to avoid lock contention.
I am not sure about the elegance of a single point of failure... if the one big machine goes down, the whole system is offline, whereas a cluster can usually survive a node crash.
The elegance falls apart when you hit the ceiling of available hardware. Getting budget approval for one massive, specialized server is often harder than adding a few cheap off-the-shelf nodes.
x168n instances make this viable again.
This follows the same logic as the Munal OS approach to memory. By removing the abstraction layers that traditionally manage distribution, you reduce the total number of failure points in the system.
Rust eliminates the GC pauses that typically kill vertical scaling in JVM-based databases. You can actually utilize terabytes of RAM without the stop-the-world events that plague traditional heaps.
Why do we keep pretending the CAP theorem is a requirement for every single app? Most developers just want a database that doesn't require a PhD in consensus algorithms to keep it from splitting.