Backblaze performance through the pool?

[JasonC]

Is anyone else out there using Backblaze to do their backups?

I've been experimenting with it, trying to maximize my backup performance, and it seems to be quite bad when using the pool as the backup source. Individual files are backing up at a decent speed, but there seems to be some kind of latency between small and moderate sized files which really adds up over time. Backblaze thinks it can do between 2-3TB a day, from the raw speed, but in practice, it's not approaching that at all. I'm wondering what other peoples experience is with this.

I'm currently testing mounting the drive to a letter and directly backing it up, and this so far seems to be doing significantly better, it's a little unfortunate, because I have 14 disks in my pool currently. Mounting to directories doesn't help, because Backblaze doesn't follow reparse points from what I've seen(Which is how disks are mounted to folders work), so I will have to also mount to disk letters if that's the only way to get optimal performance out of Backblaze.

I've of course seen the arguements this is better anyway, since you are likely to only have to replace a single disk, and identifying what's missing in a backup for a restore will be more difficult if you are choosing across the whole pool rather than an individual disk. I think I worked that problem out using Everything to index at the mount point level, but it's a moot point if I need to use driver letters to get reasonable performance out.

Anyway, I wondered what other people's experiences with using Backblaze might be, and if anyone knows how to optimize the configuration so as to get good performance out of Backblaze+DrivePool.

While I'm on the subject, I am not clear if I should have DrivePool bypass filters or not while using Backblaze...I've read the examples of where you'd maybe want to/not want to do so, but I've still not wrapped my head around it well enough to decide if this could be problematic for DrivePool or not. I suspect though, that it probably doesn't matter right now, since I'm leaning towards just hitting the individual disks directly and excluding the pool disk.

Thanks!