Continuing with the mirrors is the path I'm on. If I follow, zfs will distribute to each vdev more or less equally, so the 2TB drives will fill up and the 8TB drives will be mostly empty until they take over the load.
I'm not sure I follow why a 3 disk mirror is that much of an improvement over a 2 disk mirror (assuming one has backups), or for that matter why z2 is better than a mirror. Mostly because I don't know enough about zfs, and my math skills are deeply non-existent.
The scrub process in either event should take care of catching minor errors, cosmic rays and bitrot, I think, so it is more a question of catastrophic failure. If one assumes that there is a 2% chance that a drive will fail in the course of a year, then the chance that both of the drives will fail in a year should be around .04%, and three of three drives effectively non-existent. (Appreciating that the chance of a drive failing increase with age, so not quite.) Since you are likely to be able to replace the failed drive and resilver inside of two weeks, the chance that the second drive will fail in that period is fairly small, and the third disk doesn't seem to add much. I would be more inclined to have a single 8TB drive on the shelf available to plug in than have it spinning in a place where it might not be needed, and skip the 6TB entirely. (Yes: that 8TB being a USB drive I use for a cold backup, that I shuck if necessary to use in the pool.) If both the drives fail, then the pool fails, but it is unlikely the backup would fail too.
Similarly with a four drive z2, one has two data drives and two parity drives, or at least that is what all the explanations say. I follow if one parity drive fails, the vdev and pool can continue, but if two parity drives fail, then I guess you can read from the pool, but can you write? I guess you haven't lost data. But suppose a data drive fails? You can reconstruct the data with a replacement drive, but it seems until then both read and write are halted. Suppose the second data drive fails? Can it be reconstructed, and if not, isn't it no better than the mirror path?
I'm not sure I follow why a 3 disk mirror is that much of an improvement over a 2 disk mirror (assuming one has backups), or for that matter why z2 is better than a mirror. Mostly because I don't know enough about zfs, and my math skills are deeply non-existent.
The scrub process in either event should take care of catching minor errors, cosmic rays and bitrot, I think, so it is more a question of catastrophic failure. If one assumes that there is a 2% chance that a drive will fail in the course of a year, then the chance that both of the drives will fail in a year should be around .04%, and three of three drives effectively non-existent. (Appreciating that the chance of a drive failing increase with age, so not quite.) Since you are likely to be able to replace the failed drive and resilver inside of two weeks, the chance that the second drive will fail in that period is fairly small, and the third disk doesn't seem to add much. I would be more inclined to have a single 8TB drive on the shelf available to plug in than have it spinning in a place where it might not be needed, and skip the 6TB entirely. (Yes: that 8TB being a USB drive I use for a cold backup, that I shuck if necessary to use in the pool.) If both the drives fail, then the pool fails, but it is unlikely the backup would fail too.
Similarly with a four drive z2, one has two data drives and two parity drives, or at least that is what all the explanations say. I follow if one parity drive fails, the vdev and pool can continue, but if two parity drives fail, then I guess you can read from the pool, but can you write? I guess you haven't lost data. But suppose a data drive fails? You can reconstruct the data with a replacement drive, but it seems until then both read and write are halted. Suppose the second data drive fails? Can it be reconstructed, and if not, isn't it no better than the mirror path?
