Tradeoffs

“Object storage native” describes the new category of data systems that use the durability and reliability of object storage along with transactional capabilities like conditional writes to power higher-level semantics on top of those primitives.

That said, those same primitives do come with some latency and concurrency constraints. Notably, writing and reading object contents (PUT/GET) has a latency of around ~200ms for small objects which, if not carefully designed around, can severely impact performance. Also, on some platforms (Google Cloud and Cloudflare), compare-and-swap (CAS) features have a rate limit of once per second.

Combined, these limitations translate into two notable LoonFS limitations:

• Cold (uncached) reads may cost ~200ms of upfront latency.
• Writes in busy namespaces on some cloud providers can take up to ~1s before durable acknowledgement.

We’ve accepted these tradeoffs for the scalability and efficiency benefits, and anticipate many high-throughput file-storage use cases that do not require lower response latency. That said, this tradeoff may not be acceptable for some use cases.