ZFS Allocating Drive Bays

J

jcw

I have eight drive bays run by a Dell PERC controller. I am thinking about how to allocate them. I plan for one drive to be dedicated to swap. I plan on using one bay as a removable media for back ups. I am thinking I might use one bay as a fast ARC with a solid state media. I'm pretty sure I want to run raid-z.

I plan to run lots of jails. One of those jails will run Postgres if the database use case is a consideration in allocation. It's a hobby, but I am LARPing the professional IT guy.

I've only ever used ZFS with a single disk. I've read Lucas' book but that doesn't mean that I have a full appreciation for the practical considerations of using ZFS. The last time I touched a RAID was the late great dotcom era, and my RAID never had a problem, so I never really learned about it.

What is a new RAID/ZFS user to do with 8 drive bays?
 
I would take 1 bay for backup, as you say, That leaves 7 disks.

I would partition them so that each drive contributes to swapspace and then run geom raid1 for swap block devices.

The rest goes into a raidz2 or raidz3.


If you want more performance and more risk and less disk space you can take 6 disks and do a 3-way stripe over 3 2-way mirrors.
 
yes... you can do with zfs snapshots... and some network routing... actually i also have that configuration... running nginx, mysql, php... etc in each jail. so the main system still clean. just make sure if you gonna use certificates or ftp services with certificates... need extra configuration but its not complicated...
 
cracauer@: I added a smaller 8.0g swap partition on each drive. I added all 6 swap partitions to /etc/fstab.

From tuning(7): Keeping the swap partitions near the same size will allow the kernel to optimally stripe swap space across the N disks.

You mention above doing some geom treatment for the swap on multiple drives. It sounds like the kernel handles load sharing multiple swap partition with no geom based interventions by the admin. Why the geom mirror?

I should also mention that abandoning the dedicate swap bay/drive gives me back a bay for other uses. Thanks for the idea.
 
RaidZ. Because, that's why. :)

But seriously... I don't think I care about read or write performance. I only want to learn about high availability types of considerations. I don't want to spend double or quadruple money on stripes, mirrors, or striped mirrors.

So RaidZ let's me pretend a drive has failed, pull it, rebuild the array and see how badly I got it into a cocked hat when it's all said and done.
 
jcw said:
cracauer@: I added a smaller 8.0g swap partition on each drive. I added all 6 swap partitions to /etc/fstab.

From tuning(7): Keeping the swap partitions near the same size will allow the kernel to optimally stripe swap space across the N disks.

You mention above doing some geom treatment for the swap on multiple drives. It sounds like the kernel handles load sharing multiple swap partition with no geom based interventions by the admin. Why the geom mirror?

I should also mention that abandoning the dedicate swap bay/drive gives me back a bay for other uses. Thanks for the idea.
Click to expand...

I suggested putting the swap partitions into several raid1 so that you can hot-swap broken drives without having to reboot.
 
im not an expert but i think is not good idea to mix raid disks with swap configuration each other... i think only one swap configuration is ok... all others is mirroring... let me know if im wrong... you also have super pages or huge pages take a look.
 
Indeed.
I would also put ,
/ on RAID1
/usr/local and all data on RAIDZ

On filesystems I have / using ufs ; /usr & /var using zfs
About boot partition I have no knowledge ...
 
It would help to know about your intended work load. The more writing you intend to do, the more you should lean towards three 2-way striped mirrors. If you go that way, the strategy for the remaining disks needs some thought. But if reading is the dominant activity (it usually is), the RAIDZ2 or 3 indicated.
 
Alain De Vos said:
If I'm correct , you can do a RAID1 of a RAIDZ.
And a RAIDZ of a RAID1 ?
Click to expand...
If by raid1 you mean ZFS mirror then yes, this is pretty common in large arrays.

If you mean geom raid1 on top or raidz, that doesn't make sense because a raidz doesn't (directly) result in a block device.
 
You must log in or register to reply here.
Back
Top