3 ms·
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
by vgt 3y ago
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/ https://motherduck.com/blog/the-simple-joys-of-scaling-up/
[1] https://motherduck.com/blog/big-data-is-dead/ https://motherduck.com/blog/big-data-is-dead/
- zX41ZdbW 3y agoAttempting 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.
- riku_iki 3y ago> 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.
- dunefox 3y agoAs mentioned in another comment: developers of Clickhouse speaking ill of competitors without disclosing their bias does not convince me of Clickhouse.
- riku_iki 3y ago> in our past lives building Exabyte-scale systems. so, how did you scale to exabyte with one server?..
- vgt 3y agosee, we're not targeting these use cases at all :)