A closer look at the SpacetimeDB v2 benc...
Now that the dust has settled some more thourough investigations have been done and indeed, the benchmarks were quite misleading.
Someone achieved the same throughoput with SQLite + node in a true apples-to-apples comparison
https://foo.community/blog/spacetimedb-v2-benchmark-sqlite/
to sum up the differences:
- giving stdb a rust client, others used node clients
- stdb calls were batched forcing a cpu bottleneck instead of a disk one
- drizzle ORM wasn’t even being used correctly
- they were using journal_mode=delete which slows down SQLite tremendously
- using websockets for their own database, http transport + json serde for others
It’s still a cool piece of tech, but I wish they had just presented it in a more honest light.
Someone achieved the same throughoput with SQLite + node in a true apples-to-apples comparison
https://foo.community/blog/spacetimedb-v2-benchmark-sqlite/
to sum up the differences:
- giving stdb a rust client, others used node clients
- stdb calls were batched forcing a cpu bottleneck instead of a disk one
- drizzle ORM wasn’t even being used correctly
- they were using journal_mode=delete which slows down SQLite tremendously
- using websockets for their own database, http transport + json serde for others
It’s still a cool piece of tech, but I wish they had just presented it in a more honest light.
Foo Community
Unpacking the SpacetimeDB v2 benchmark: the physics of disk writes, the math behind transaction batching, and what it actually takes to push a database to its limits.
