The only difference really is single-point-of-failure. Multiple threads writing to the same DB record can be assisted with transactions or locking - but again,

The only difference really is single-point-of-failure. Multiple threads writing to the same DB record can be assisted with transactions or locking - but again, the DB is, in the end, absolutely authoritative. What it says goes, and if it's unavailable, so is everything. When solving for high Availability through a distributed cluster, you might give up immediate Consistency. But partitions can end up giving different answers (authoritatively, even) and an unavailable master record doesn't mean something won't answer as if it _were_ authoritative. A single node DB - like a MySQL server - still has failure points, but none that plague CAP systems (well, except Availability, but that's how we got into this mess to begin with).