RAM vs Disk Storage

[fhalde]

Depending on the requirements of the application, Qdrant can use one of the data storage options. The choice has to be made between the search speed and the size of the RAM used.

In-memory storage - Stores all vectors in RAM, has the highest speed since disk access is required only for persistence.

Memmap storage - Creates a virtual address space associated with the file on disk. Wiki. Mmapped files are not directly loaded into RAM. Instead, they use page cache to access the contents of the file. This scheme allows flexible use of available memory. With sufficient RAM, it is almost as fast as in-memory storage.

Perhaps it is worth addressing what does it mean for vectors to be in RAM. Does it mean the collection is not durable? Should I expect data loss? If not, what is durably persisted?

Am I missing some context here? Thanks!

[timvisee]

Everything is stored durably, specifically on your persisted storage. That means that if you'd cut the power to your machine, no data is lost. So nothing to worry about there!

When we say data is in memory, we actually mean that it's both on disk and in memory. It being in memory is significant for fast performance, but it comes with more memory usage.

I this could benefit from clarifying that in our documentation.

[fhalde]

Thanks @timvisee – indeed it has to be durable ;)

[fhalde]

another confusion we had was, what does it mean to set on_disk: false

[timvisee]

what does it mean to set on_disk: false

Data is still on disk, but also always in memory (RAM).

[fhalde]

do you have any documentation or YT video explaining the internals of Qdrant? happy to take a look on how everything comes together

[timvisee]

We don't really document our internals publicly because it changes all the time.

I can recommend our free course available here: https://qdrant.tech/course/
Though that doesn't really go into the internals.

[fhalde]

nevermind

on_disk: true - The index is stored on disk, which lets you save memory. This may slow down search performance.
on_disk: false - The index is still persisted on disk, but it is also loaded into memory for faster search.