I'm pretty sure you have to use rsync.
Oops meant Grsync.
1 Like
there are two, on zfs, and the first one is mostly common sense. the second one has the good stuff. they have a whole series if you're into bsd. I could see all the arch folks liking bsd pretty well though.
Loved the video. I hope there will be more like it!
1 Like
Just awesome. Needs to be an annual thing, no doubt.
A video on how to encorporate ZFS into a linux desktop with a snapshot OS would be amazing. I have wanted to do that for ever!!
I think they are working on zfsonlinux being able to be used on root. I think you can actually do it now but they don't recommend it yet.
So Allan Jude is responsible for hosting tons of video content online...... I don't think its netflix soo......
very nice video! thx for that!
Awesome video. Glad to see more on zfs, been looking into rebuilding a server and zfs seems to suit my needs.
I still want BTRFS to become stable. To be honest I'd rather move my NAS to BTRFS raid 10 than ZFS for a few reasons. ZFS on Linux just seems clunky because of the 3rd party kernel module and SPL. But for now I'll just wait and see how BTRFS is in the next few kernel releases. I think it's stable right now as long as you don't use raid 5 or 6 but we'll see...hopefully that'll change soon.
Modern storage systems test these batteries themselves. Most enterprise servers report when BBU or SuperCap require replacement. Even if somehow your server can't do this by itself, it's easily automated, at least for LSI controllers. It's actually more troublesome when system does it all by itself because more often than not you have no control over the scheduler for this stuff, which means sudden unexpected write-through mode on half of your LUNs.
They also have flash + supercapacitor now -- no battery testing at all.
When talking about lots of disks/shelves, it is still a huge mess though. Seen plenty of shorn writes on "hardware" arrays even with super capacitors.
Supercaps are still tested from time to time, afaik. And they may have their own problems (CacheVault's gas gauge issue, for example, may screw its learning cycle and make it unreliable).
@wendell, since you're here, can you explain what was that about not storing metadata on adjacent disks? How the hell is it even possible?
the disk shelves report a lot of information about the arrangement of disks to the host os. if you have a bunch of disks in a server, not so much, but commercial grade stuff has redundant disk connections. each disk has at least two totally isolated paths to it and typically those paths are switched into a more complicated fabric (so more than two paths until you are local, then two paths).
Think of it like switched ethernet, except every disk has two ethernet ports into two separate ethernet switches. each with at least two uplinks elsewhere. That's modern disk stuff.
1 Like
Yes, I've read (well, skimmed through) the manual on SAS (I guess you read the same one, judging by analogy you've brought up). The topology isn't exactly relevant to the question. But this:
is largely dependent on the shelf itself, on its backplane, on drive interfaces and shelf form factor. You may have 1x12, 1x24, 3x4 slot placement. In some SunFire servers, when they switched to 2.5" drives, I think there were 2 drives in a first row then 4 or 6 drives in a second row of slots. There are shelves which won't report the "slot(X)" stuff to mpt2sas, and all you're left with is "phy(Y)" at best - which doesn't correlate to a slot number. And if it's SCSI, not SAS - then I don't think you have the slot number reported anywhere. It's hard to put in my head that zfs devs actually bothered to add extra calculations which would work properly only in some specific cases and won't do much (or may even do exactly the opposite to their announced purpose) otherwise.
Dont forget worldwide numbers -- lot of info encoded in there as well. I wonder if it takes into account adjacent worldwide numbers. That'd be easy to do that as well.
With a hundred+ QA engineers working on the solaris side, I'd say it does exactly what he says on that type of hardware. Anything else might be a crapshoot, though.
I doubt there's anything encoded in wwn, except vendor id, maybe model and serial number. I'm pretty sure drive wwn doesn't change if you install it into a different slot.
But I have a stupid idea. I wonder if it's possible to indirectly obtain the information about how drives are located relative to each other, say, by writing to one drive and reading SMART of other drives (error rates and vibration sensors) at the same time.
Wonderfully informative and enjoyable video. But... ZFS on a laptop? Please, explain yourselves! I'd really like to have a mature COW filesystem on my laptop, my desktop and everything but I was under the impression that one should not run ZFS without boatloads of spinning rust and some ECC RAM.
(My computers typically have as much RAM as I can get my hands on, but only two disks.)
Wendell, how to the ex-google server going with Fedora and ZFS for you guys ? Is there a follow up video in it ?
1 Like
There are about 3 followups in varying states of completion 
it's doing pretty well so far knock on wood.
4 Likes