SSD zpool

[Youri Andropov]

Hi there,

Is it technically possible to use an array of SSD, let's say 6x 1 TB, and use them as normal HDD in a RAID Z1 (or Z2) pool ?
What about trim ?

I'm thinking about it because of the failure rate of normal HDD.

 

[Ericloewe]

Hi there,

Is it technically possible to use an array of SSD, let's say 6x 1 TB, and use them as normal HDD in a RAID Z1 (or Z2) pool ?
What about trim ?

I'm thinking about it because of the failure rate of normal HDD.

It's certainly possible to use SSDs. I believe TRIM is supported, but I recommend you double-check.

But don't use SSDs because you're afraid of hard drive failure. They're unlikely to fail less.

 

[anodos]

An array of SSDs is good if you need IOPs or are running 10 gigabit. If I were concerned about MTBF I would go with SAS drives. Although if you're in a place where you can't easily control temp or vibration an SSD array might be a good option.

 

[joeschmuck]

I think one of the developers might be able to answer if TRIM is supported however I don't feel it would be an issue speed wise as it will still write faster than a spinning hard drive even without TRIM. And of course you could build the FreeNAS using SSDs.

The longevity of a SSD depends on a few factors if you eliminate infant mortality. One is how many write operations can be performed to the SSD before it starts to affect overall drive capacity as the cells wear out. Wear leveling minimizes this effect however it does exist. But still you should look into how long a typical SSD of the type you want to purchase could last if it were written to constantly.

And from what I've read, the best hard drives are equal to current SSD's (with TRIM support) with respect to longevity, but I suspect that will change in the near future.

Now if you are trying to build a very compact, quiet, and fast NAS, sure the SSD's would be fine. Could you toss me four 1TB SSD's as well while you're at it? I can't afford it myself.;)

 

[cyberjock]

TRIM is only supported on the ZIL or L2ARC(I forget which).

 

[Youri Andropov]

If I were concerned about MTBF I would go with SAS drives.

The few enterprise-class HDD I've looked have a MTBF of about 1.4 M hours (Seagate Constellation ES for example). Samsung SSD 840 EVO 1 TB MTBF is 1.5 M hours.
Most of my NAS activity is reading, so I expect SSD to live long.

 

[Ericloewe]

The few enterprise-class HDD I've looked have a MTBF of about 1.4 M hours (Seagate Constellation ES for example). Samsung SSD 840 EVO 1 TB MTBF is 1.5 M hours.
Most of my NAS activity is reading, so I expect SSD to live long.

Don't trust MTBF numbers, they're mostly drawn out of somebody's ass.

 

[cyberjock]

Don't trust MTBF numbers, they're mostly drawn out of somebody's ass.

And not just anybody's either! An engineer's! Yuck!

 

[TheSmoker]

Trim on zfs ssds is supported starting with freebsd 9.2 and it is enabled by default.
Freenas 9.2.1.3 is already on freebsd 9.2p3. So you will be ok.

 

[cyberjock]

Trim on zfs ssds is supported starting with freebsd 9.2 and it is enabled by default.
Freenas 9.2.1.3 is already on freebsd 9.2p3. So you will be ok.

I'm pretty sure there's some kind of bug related to TRIM and ZFS and it doesn't work quite how it was documented. But I don't know more than that unfortunately. As very few people run full SSD pools TRIM is a minor concern in the big reality of the situation. ;)

 

[Youri Andropov]

TRIM seems supported : see http://www.freebsd.org/releases/9.2R/relnotes.html

Code:

# sysctl vfs.zfs.trim
vfs.zfs.trim.enabled: 1
vfs.zfs.trim.max_interval: 1
vfs.zfs.trim.timeout: 30
vfs.zfs.trim.txg_delay: 32

 

[Binary Buddha]

TRIM seems supported : see http://www.freebsd.org/releases/9.2R/relnotes.html

Code:

# sysctl vfs.zfs.trim
vfs.zfs.trim.enabled: 1
vfs.zfs.trim.max_interval: 1
vfs.zfs.trim.timeout: 30
vfs.zfs.trim.txg_delay: 32

But, as Cyberjock stated... it could be buggy.

 

[Mirfster]

Binary Buddha wins the "Thread Necro'er" Award for the day... :P

 

[Stux]

TRIM is only supported on the ZIL or L2ARC(I forget which).

I wouldn't expect lack of trim to be a big deal on a CoW FS like ZFS, as long as the SSD had decent garbage collection.

 

[HoneyBadger]

As long as we're necro'ing this thread;

TRIM works in its current implementation, what you have to watch for is the "TRIM on init" issue when adding SSD-based vdevs to a pool - TRIM seems to currently be issued as "wait for completion" and will actually pause pool I/O, which can cause timeouts/VM stun/other nasty business. Also affects addition of L2ARC/SLOG.

I wouldn't expect lack of trim to be a big deal on a CoW FS like ZFS, as long as the SSD had decent garbage collection.

TRIM will always help garbage collection. There's a pretty good writeup on Ars about it.

 

[Ericloewe]

It's even more important on a COW filesystem, because every single operation will write a new block. Running garbage collection without TRIM would be a nightmare for any drive with less-than-spectacular overprovisioning - which is nearly all SSDs.

Of course, COW considerations don't apply to L2ARC and SLOG. Additionally, L2ARC is rarely written to and the ZIL is tiny by modern SSD standards, so TRIM is not an absolute requirement.