“Even with old hard drives” - does that mean spinning platters? IMO, we should optimize around the characteristics of SSDs. SSDs are more than just “HDDs but faster”.
It occurs to me that there’s a rift between “optimized for retro computers” and “optimized for low-spec computers”. I’m more interested in the latter, TBH.
Optimizing for cheap, low-speed SSDs is roughly the same as optimizing for hard drives though. They both love writing in large blocks, hard drives because they spin that way, SSDs because with large blocks they need to care less about garbage collection.
In fact, one of the NVMe extensions is Zoned Namespaces, which is building on the base that was used for host-aware SMR drives - writing blocks of data linearly is more efficient, because the SSD firmware doesn’t need to handle any garbage collection and doesn’t need any over-provisioning.
“Even with old hard drives” - does that mean spinning platters? IMO, we should optimize around the characteristics of SSDs. SSDs are more than just “HDDs but faster”.
It occurs to me that there’s a rift between “optimized for retro computers” and “optimized for low-spec computers”. I’m more interested in the latter, TBH.
Designing software for cheap eMMCs and SD cards! I like it.
Optimizing for cheap, low-speed SSDs is roughly the same as optimizing for hard drives though. They both love writing in large blocks, hard drives because they spin that way, SSDs because with large blocks they need to care less about garbage collection.
In fact, one of the NVMe extensions is Zoned Namespaces, which is building on the base that was used for host-aware SMR drives - writing blocks of data linearly is more efficient, because the SSD firmware doesn’t need to handle any garbage collection and doesn’t need any over-provisioning.