What are Borg's I/O patterns and its compatibility with SMR/tape storage devices?

[ttimasdf]

Hi everyone! I'm new to Borg and interested in using it to back up my files to a NAS.

Initially, I assumed that Borg performs incremental backups in a write-and-append manner, which works well for SMR storage where in-place modifications can be costly.

I've gone through the Data Structures and File Formats section, but I still have some questions about certain details.

Are these object files accessed during the execution of borg create? Considering that SMR disks perform poorly with random file access and files are divided into small chunks, I assume that borg extract and borg compact would be significantly slow. Does the same apply to borg create as well?

[ThomasWaldmann]

borg (and its predecessor attic) were not designed for tapes, but for random access media like HDDs.

but (afaik) there are no major issues with SMR HDDs. Considering that most of 2.5" USB HDDs these days use SMR, we would know.

while borg create is running, most I/O is just appending new data to the current segment file until it is "full" (default: 500MB) and then continuing with the next segment file.

borg extract could be a bit more random I/O, pulling chunks from the repo in the order they are referenced from the metadata.

borg compact processes each non-compact segment file sequentially, shuffling still needed data from it to a new segment file.

What I wrote above describes borg 1.x versions, which are the current, stable production releases.

Unreleased code from master branch works very differently right now, but it is still to early to say how the borg2 release will behave.