SSDs as main storage

[joeschmuck]

You can use AMD hardware, I have a system but to be honest (see tagline), if I could do it all over again, I really want IPMI and the AMD boards don't have that. Also there is no documentation on how the AMD board notifies you if the ECC RAM has an issue. My system is virtually silent, the AMD CPU fan is small and you can't hear it. I mounted everything in a large case so air flow is great and ultra quiet fans can be used. The hard drives make virtually no noise as well. I would say my system is ultra quiet, but it's not small. The cost was low as well (I'm a cheap bastard). I now desire to upgrade two systems, my FreeNAS and my Sophos but the costs are more than I want to spend right now. The holidays are hopefully going to be good to me so I can purchase some stuff on sale.

As for using SSD's as a FreeNAS storage medium, I wouldn't unless you purchased the high end enterprise SSD's (SLC) due to wear. Either way, SSD's are not cost effective unless you are a business with high availability requirements.

And there is nothing wrong with your old FreeNAS system, sure it's old but hey, I started there as well. As others have suggested, install an Intel Gbit NIC into the device and that will speed up your backups significantly, provided all your network connections are Gbit speeds (no 10/100Mbit hub/switch).

 

[Something]

SLC SSDs would be beyond excessive. Especially as mainstream SSDs of not OCZ quality can handle multiple hundred TBs to Petabytes of writes.

http://techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead

Like I was kinda saying though, the speeds of an SSD would be great since large capacity isn't needed, but if large capacity isn't needed an SSD would be speeding up already fairly quick backups.

 

[tvsjr]

Also keep in mind you would need to monitor the lifespan indicators on the SSDs... you can read/write spinning rust a lot more times than you can an SSD (although the gap is narrowing, assuming you get good SSDs).

 

[Something]

I think we all agree that SSDs for main storage in a consumer grade FreeNAS system just isn't viable yet then.

Too little space needed and its largely acceleration of backups that are quick already. But to have enough storage needed that SSD speeds tangibly help requires a massive investment and supporting 10GbE+ infrastructure.

 

[footer]

What I wonder though is if you should bother with a FreeNAS box at all if your storage needs are so low. You could consider a simple NAS, if not just an external HD. Alongside a cloud backup of some kind.

This is a good question and a very good point. However, even a simple NAS (off the shelf) is relatively expensive compared to rolling your own, not to mention limited flexibility down the road. And as for external HDs, I'm just not comfortable with those and I honestly don't think they're reliable long term.

But the real beauty of FreeNAS is sharing amongst devices on the home network. I'm sure there are simpler and maybe cheaper alternatives, but a tinkering geek like me finds much more satisfaction in rolling my own FreeNAS!

I read through the Hardware recommendations (read this first) thread and did some quick research and going on the cheap, can get a Supermicro mobo, 16GB of ECC RAM, and Intel processor for around $350. Granted these are all on the lower priced side but they would get me by for my meager needs and take me well into the future. I could reuse my case and drives and be running the latest FreeNAS with minimal expense and have a relatively new/reliable system that would only need new drives in the future (spinning or SSD) to increase space.

Again, I really appreciate the discussion and information this thread has provided. I'll keep watching the SSD landscape as it evolves but based on all the feedback here so far, it's not prime time yet as an alternative for a FreeNAS box!

 

[Handle]

Another viable use case for SSDs are energy savings in countries where the costs of electricity are quite high. In Germany the cost per kWh is around 26¢ so depending on the scenario it might pay for itself if therefore the disks can stay in spindown most of the time.

 

[Something]

This is a good question and a very good point. However, even a simple NAS (off the shelf) is relatively expensive compared to rolling your own, not to mention limited flexibility down the road. And as for external HDs, I'm just not comfortable with those and I honestly don't think they're reliable long term.

Your space needs seem incredibly thin though. Which is why I recommend a simple NAS. You don't seem to need expansion, if under a terabyte is plenty. Even a 2x4TB NAS would be plenty for you it seems.

You are right though, rolling your own is cheaper. And you can still roll your own and end up with a simple NAS.

But the real beauty of FreeNAS is sharing amongst devices on the home network. I'm sure there are simpler and maybe cheaper alternatives, but a tinkering geek like me finds much more satisfaction in rolling my own FreeNAS!

There aren't (as far as I know), FreeNAS/Nas4Free or bust!

I read through the Hardware recommendations (read this first) thread and did some quick research and going on the cheap, can get a Supermicro mobo, 16GB of ECC RAM, and Intel processor for around $350.

You could consider the Atom SoC boards.

Granted these are all on the lower priced side but they would get me by for my meager needs and take me well into the future. I could reuse my case and drives and be running the latest FreeNAS with minimal expense and have a relatively new/reliable system that would only need new drives in the future (spinning or SSD) to increase space.

Given your meager needs, HDs would be more recommended for a long time coming.

 

[diskdiddler]

Sooner or later, 2,4,8,12TB SSD's will come down - and if rumours are to be believed, surprisingly soon.
Eventually FreeNAS needs to be prepared for users trying to use it in an SSD only environment.

Is it even possible as it stands and if not, how far away is it?

 

[jgreco]

Once you look at the density of a device like the M.2 Samsung 950 Pro, it's obvious that this is possible even today, and that it is just a matter of playing Tetris with silicon chips to get things to fit. For servers, a 15mm high 2.5" drive is perfectly acceptable.

We're already at a point that if cost wasn't an issue, you could build immense capacity 3.5" SSD's and blow away the density of HDD. Again, all you really need to do is contemplate the amazing density that we've already reached with M.2.

SSD based pools are nothing new, but the truth is that ZFS isn't particularly optimized for that use model, so as someone else said, the endurance characteristics will need to continue to improve. I've already got an all-SSD pool that I intend to expand.

 

[Something]

Sooner or later, 2,4,8,12TB SSD's will come down - and if rumours are to be believed, surprisingly soon.
Eventually FreeNAS needs to be prepared for users trying to use it in an SSD only environment.

Is it even possible as it stands and if not, how far away is it?

SSDs don't exist in a vacuum, and for most, having high performance, highly efficient drives of sufficient storage capacity wouldn't be worth the price of admission unless they're doing things on the FreeNAS machine that requires internal throughput be incredibly high or the network being sufficient in speed to saturate those SSDs.

With RAID0 850 EVOs I was pushing 1GBps, which means if storage space was sufficient, I could transfer 3.5TBph. That'd require SSD running local machines on 10GbE to make such transfers possible. And as of yet both 10GbE and SSD pricing seems a bit far out.

It is entirely possible at this point but unless you have enterprise style, insane storage speed demands SSD based NASes seem a solution to a problem that doesn't exist. As you can't compete in capacity that well or probably just can't afford it. Because you're accelerating what fundamentally amounts to small transfers. Is it really worth paying all that much to speed smaller backups up I don't think so.

Going forward though, 10GbE will only cheapen and SSDs are continuing to drop like a rock. I think within the next 3 years for the heavy hitter 'enthusiasts' of the homelab and FreeNAS community will strongly be considering SSD-only, like how Intel made NVMe SSDs heavy enthusiast affordable with the 750 and Samsung made PCIe SSDs heavy enthusiast affordable with PM951. And with the price drops on the 750 and the newly released 950 Pro from Samsung, PCIe NVMe SSDs have become enthusiast affordable, the equivalent here being another 2 years, so 2020. SSDs don't need to reach price parity with HDs to overtake them, as they steal into capacity and pricing better reaches HDs they're done. The storage trifecta is capacity, speed and pricing. Capacity is being reached, speed is a blowout but pricing remain firmly in HDs favor.

Currently the cheapest HDs in GBs per dollar match where they were about half a decade ago around 0.02X dollars per GB. SSDs have come down an unholy amount in that timespan, and they're more or less following, if not exceeding, Moore's observation of half the price every year and a half, two years. While they're currently a full order of magnitude 10x the dollars per GB, if they continue at the present rate they'll halve three times in the next 3-5 years. Which would put them on price parity with current HDs.

HDs can continue to hold the lead in pricing in this time, but unless storage needs rise dramatically, SSDs will be heavily favored by pricing.

 

[jgreco]

SSDs don't exist in a vacuum, and for most, having high performance, highly efficient drives of sufficient storage capacity wouldn't be worth the price of admission unless they're doing things on the FreeNAS machine that requires internal throughput be incredibly high or the network being sufficient in speed to saturate those SSDs.

With apologies, you've actually missed out on the thing that consistently drives storage guys to drink: seek times. You don't need to need "incredibly high" internal throughput or to "saturate [...] SSDs" to find that SSD storage is leaps and bounds more awesome in terms of the significantly reduced seek speeds; a hard disk can have a rough time doing more than ~100 IOPS but an SSD can usually do at least 10x times that without a problem. This is particularly awesome for things such as small file (think: document) storage or VM/database block storage. You don't need to be pushing an SSD to anywhere near its throughput limit, and it will still be head and shoulders above the HDD.