Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by didip | Hacker News Reader

Parent Full thread

didip·

Does Motherduck have distributed storage, query, and caching layers?

View on HN

For compute specifically MotherDuck is a scale-up, not scale-out system [0]

Storage is naturally distributed due to underlying IaaS.

Rather, that seems to be an implementation detail. What use cases are relevant is the key. Are we chasing big Apple-sized workloads? No. [1]

That said, scaling-up on EC2, especially with our one-instance-per-user architecture, fits vast majority of workloads we've seen in our past lives building Exabyte-scale systems.

[0] https://motherduck.com/blog/the-simple-joys-of-scaling-up/ [1] https://motherduck.com/blog/big-data-is-dead/

Attempting to present the inability to scale out as a sort of advantage sounds strange to me. The problem with duckdb is that it does not work as a database server. Occasional data analysis may be more or less ok, but you don't need duckdb for that - you can use proven, boring open-source technologies, such as ClickHouse.

> Attempting to present the inability to scale out as a sort of advantage sounds strange to me.

it is a scope, not inability. Building single node system is much easier than distributed for fast complex algos, meaning team can ship more features.

> boring open-source technologies, such as ClickHouse.

to me ClickHouse is not boring, it is fragile and failing all the time with OOMs in various places.

As mentioned in another comment: developers of Clickhouse speaking ill of competitors without disclosing their bias does not convince me of Clickhouse.

> in our past lives building Exabyte-scale systems.

so, how did you scale to exabyte with one server?..

see, we're not targeting these use cases at all :)

Reply on news.ycombinator.com