Post
@psyhackological@Tubsta yes, it is. Mastodon storage contains a lot (A LOT!) of small files and SeaweedFS is tuned for this use case, too.
@stefano Did you just follow the single node setup from here? https://github.com/seaweedfs/seaweedfs/wiki/Production-Setup I'd be interested in the detailed specifics on how you set up your storage. I'm looking at a single node S3 storage using a 2TB NVMe drive in a m930q tiny (I have heaps of these things).
@psyhackological@Tubsta I conducted some experimental tests. I set up Garage and SeaweedFS on the same machine - with rotating disks - the same one on which Minio runs. I then used rclone to copy files from Minio (which was in production) to Garage and SeaweedFS. I repeated the test several times.
With Garage, after a few GB, the process slows down, the CPU load increases (and by a lot), and the I/O wait becomes high. After about 80GB, I have to stop; the load becomes extremely high, and the latencies are monstrous.
With SeaweedFS, the load remains constant, and the performance is the same. The I/O wait remains minimal.
I've attached a graph of the machine's CPU load. On the left is when I performed the rclone to Garage. On the right, to SeaweedFS. Note that the bulk of the load in the second part is caused by Minio's reads.
With Garage, after a few GB, the process slows down, the CPU load increases (and by a lot), and the I/O wait becomes high. After about 80GB, I have to stop; the load becomes extremely high, and the latencies are monstrous.
With SeaweedFS, the load remains constant, and the performance is the same. The I/O wait remains minimal.
I've attached a graph of the machine's CPU load. On the left is when I performed the rclone to Garage. On the right, to SeaweedFS. Note that the bulk of the load in the second part is caused by Minio's reads.
@psyhackological@Tubsta Probably Garage isn't tuned for millions of small files, while SeaweedFS is. Or it's just more mature - SeaweedFS has been around for more than a decade.
1 more replies (not shown)
