After suffering with some hardware-induced kernel panics on my 27” iMac, an upgrade was warranted. Its internal storage is magnetic, and its external storage was all questionably reliable FW800 that was near its capacity. Being an iMac, I thus had to get a computer, new screen, and new storage.
I wanted to wait for the new Mac Pro, but there are two problems with that approach. The first is that it’s not out yet. The second is that it’s going to be expensive. That said, I don’t really need high-end graphics and Xeons. All that’s essential is decent I/O performance.
Mercifully, the Mac Mini is now available with both solid state internal disk and a thunderbolt port (not the 6 of the Mac Pro, but I’ll take it). Upside is, the approach I’ve taken means a Mac Pro is a drop-in-place replacement, if and when I do go that route.
My original plan was to have a small solid state stripe (fast random scratch) and a large magnetic stripe (fast sequential and/or large scratch) over thunderbolt. I would then keep the majority of the non-immediate storage on a large Synology NAS. With link aggregation, network attached storage is more than fast enough for my secondary storage needs.
There’s a problem with a big NAS, though. Offsite backup is tricky and expensive. Most of the consumer offsite providers don’t support NAS backup. There are means to hack this together but they don’t appear to be supported or reliable. The closest thing to being viable and affordable is Amazon’s tape-based Glacier product.
Unfortunately, Glacier has two cost structures to worry about. The first is transactions, and the second is peak recovery rates. In looking through my data flows, it was clear I would have a high transaction count. No good. Worse, recovery could run into absurdly expensive rates depending on when it was done. Also not good.
By comparison, if I just plug all the disks into the mac and use Backblaze, it’s $5 a month. So I did that, and will continue to saturate my upstream for a couple of months until that finishes. That looks like this, pretty much 24x7:
Backblaze deserves some credit. I can push about 40 GB/day to them at these rates. This is off my theoretical upstream capability (around 35 Mbps), but still worlds better than most of the competition. One could do a lot worse than 10 Mbps being the limit to keeping their offsite synchronized (that’ll keep me up to date faster than Time Machine, provided I haven’t just slurped down a couple of 32GB memory cards).
So yeah, storage. I have two RAID-0 8TB stripes that are used for persistent storage. These are synchronized (hot and warm) nightly via launchd/rsync. Same idea with solid state scratch at 220GB a piece. All of that is connected via Thunderbolt. Local backup is via Time Machine to a USB3 12TB RAID-5 volume. It’s not elegant or attractive, but it works. It will be a while before I fill up those disks.
The daisy-chained Thunderbolt solution isn’t fantastic, since there are only two channels, 10Gbps in, and 10Gbps out. This is an area where I’d rather they removed the FW800 port from the Mini and replaced it with a second Thunderbolt port. It doesn’t help that the display is also on the same daisy chain. I don’t know enough about Thunderbolt to comment intelligently on the feasibility, but it appears that all of the I/O is also not point to point, which means intra device transfers can actually dip below link aggregated gigabit ethernet rates (hard to leverage faster than 4Gbps in practice, but the fabric’s rated to 48Gbps continuous). It works out.
With all of that said, I’m able to move from Thunderbolt to mac and vice-versa at a sustained rate approaching a gigabyte per second. I’ll take it.
Nothing about the whole setup is stupendously fast, but it’s more than fast enough for my needs. Were realistic offsite NAS backup I’d rather go that route and have fewer things plugged into it. In the meantime, I would like to have more than one Thunderbolt plug, and that may ultimately be the most compelling reason to go Mac Pro in the future.