HP Microserver Gen10 - Setup options?

[0nighthawk0]

Hi all,

I've got Truenas 12 working on my HP Microserver Gen10.

This has 4 bays for 3.5" drives and i've got a 3TB WD Red (Nasware 3.0) drive in each bay.
It also has an extra SATA port internally and i have a 450GB Retail SDD in there.

I'm currently running TrueNAS from a 64GB Samsung USB stick and its pretty speedy, but i've read on the forums that booting from Flash Media isn't really supported any more, or should be avoided. The thing is i'd prefer not to waste the 450GB SDD (taking it out is an option) and I could get a much smaller SSD for a Freenas install.

There is not enough room in the case for another 3TB WD Red Drive to use as a hot spare or anything like that.

Currently, i'm running two mirrored vdev's - for a total space of 5.45GB in the pool. However open to suggestions about better use of this.

I've run some servers from USB for years as it's easy to backup config and reinstall/restore.

Might seem like a daft question, but is there any way to reduce writes to the USB other than turning some logging/atime off? Once i've got the server to a configured state, i can;t see it changing in any meaningful way other than writes to the pool and logging?

I've read some people are running from USB but with 2280 length drives in a usb style caddy - anyone got any experience of these - would anyone recommend some smaller boot/cheaper boot drives? I've looked at a few small and cheap intel options, just want ot know what works for people.

Just questioning/optimising my setup.

Thanks in advance.

 

[alberthartland]

Also running a Gen 8 Microserver with 4 x Ironwolf drives and booting off usb.

I'd love to read the posts about excessive writes under usb. I've been running my Gen7 with FreeNAS 9 -> 11 over 7 years and I don't recall any issues.

Looking forward to reading about anything I should look out for.

 

[Etorix]

With four bays, there's no better use than a stripe of mirrors. A RAIDZ2 would be slightly safer, especially if you move to larger drives, but for the same capacity and less IOPS.

The issue with USB boot devices is that they wear out. If you don't mind to reinstall when the stick fails, always have a recent backup of the configuration at hand and keep going. Otherwise, replace the 450 GB by any small SSD.

 

[0nighthawk0]

Thanks for the responses. I'm just doing some drive burn-in and tests then i'll hopefully be in a position to tesst if there is any speed increases from using SSD as cache. I may just use it as fast storage. Just trying to come up with an optimal config although obviously it depends on use case scenario.

 

[0nighthawk0]

Just out of interest if I used the SSD as an L2ARC/Read Cache vdev. If it the l2arc failed would the pool still remain working or would i have to remove the drive from the config?

 

[Etorix]

L2ARC is just a cache of data that otherwise exists in the pool. If the L2ARC dies, ZFS keeps working directly from the pool.
But operating a L2ARC requires a part of the ARC (RAM), so the common advice is to max out RAM before considering a L2ARC. If your short on RAM, a L2ARC may even slow you down.

 

[0nighthawk0]

L2ARC is just a cache of data that otherwise exists in the pool. If the L2ARC dies, ZFS keeps working directly from the pool.
But operating a L2ARC requires a part of the ARC (RAM), so the common advice is to max out RAM before considering a L2ARC. If your short on RAM, a L2ARC may even slow you down.

Thanks for that, more proof RAM is king for ZFS.

I've got my pool on Truenas up and running, i've got my old server running Nas4free/XigmaNAS with my original ZFS pool.
Just doing a scrub on the old pool atm but don't expect any problems.
I need to transfer all the data from that Pool to my new pool but don't want to keep the datasets.
I'm going to be restructuring the folders/data.
zfs/send recieve from what i recall last time i used it, copies the dataset from snapshot. While i do want consistent data i don't want to be creating datasets that I won't be using later. I was hoping to pre-setup the new datasets with new properties.

Any advice on transferring this data in a consistent way other than zfs send/recv?