Would you use consumer SSDs in a server?

[Chicken76]

Specifically, would you use any of the following:

Samsung 840 PRO
Samsung 850 PRO
Crucial MX100

I lean heavily towards the MX100s because they are cheap and for about the same price I can get one size bigger, which would help with the writes and endurance. Plus, they have capacitors, similar to the enterprise SSDs.

Also, would you pair different drives (models/manufacturer) but same size in RAID1 arrays (C224 chipset)? I'm thinking that it might save me in case of bugs one might experience in 24/7 use in consumer drive firmware.

The server would run MSSQL. The writes would be 2-5 GB per day.

 

[docp]

I would do so provided if back ups are daily

 

[Chicken76]

I would do so provided if back ups are daily

Thanks. Do you think the idea of combining different SSDs in mirror arrays has any merits?

 

[IndyColtsFan]

I'd think twice about using consumer SSDS in a server if that server were a critical production server. Lab, dev, or test server? Sure. I have 4 SSDs in my Hyper V host at home along with 8 spindles.

 

[Railgun]

It depends on your budget, criticality of the server, use case, how much data will be written over time, and so on.

If it's tons of reads, probably. If it's some DB with a lot of writes, probably not. YMMV.

 

[corkyg]

That would depend on the server. Some people use servers at home that are totally unrelated to business. They would be OK - they are entertainment, not mission critical. If it was a server in a business, backup would be essential.

 

[redzo]

If you seek maximum performance you could drop 2 x Samsung 850 pros in a mirror anytime. What is the problem if you constantly monitor the server? If one fails, you come back with a replacement. Additionally, since their specs are the same, it would be wise to monitor the TBW values. But you already backup everything somewhere on a nas just in case the entire mirror fails right? So what's the problem in using 2 consumer drives like the 850 pro in a server mirror?

Edit:
- Even small business servers should be backed-up by an ups. So I don't see the point in going for those fancy power loss protection mechanisms at ssd device level.
- The pro has higher endurance(double).
- As long as you back up everything somewhere else, I really do not see any concern at all. I mean, you would have backed-up the mechanical data also.
* I would be more concerned in if not/how passing trim commands through the server's raid controller.

 

[BonzaiDuck]

To me, it's a balance between cost per GB, speed and related factors.

I use a Sammy EVO 120GB for the WHS-11 OS/System/boot drive. I was actually backing it up to another SSD -- a 60GB unit -- but it seemed to be filling up more than I'd anticipated. So I swapped in a 320GB Blue WD drive for the automatic OS backup until I sort out what to do next.

I don't see a problem using an SSD in this way for a server. I separate backup of my drive pool of data from the backup of the OS.

 

[Carson Dyle]

Yes.

The only thing about using a so-called consumer SSD drive is how often you can expect to deal with a broken drive. Everything else is the same. You have to take exactly the same precautions in terms of assuring system availability and recovery in case of failure.

Performance differences are a moot concern, IMO, as performance is already so far ahead of anything we were doing with mechanical drives just a few years ago.

 

[Chicken76]

Thanks guys. A good UPS will be used and also daily backups are going to be performed.

Do you think using two drives from different manufacturers in a mirror is a bad idea?

 

[mrpiggy]

Depends on the usage of the server. For a heavy use database server with tons of constant writes and reads, no.. For a simple file server that is mostly static (like a document/file storage server that is mostly reads and not much writes) cheap consumer SSD's would probably be perfectly fine. As it is writes that wear out NAND, if there isn't going to be a ton of writing, then consumer SSD's can work, just make sure you have good backups and hopefully extra failover drives.

 

[redzo]

Also, would you pair different drives (models/manufacturer) but same size in RAID1 arrays (C224 chipset)? I'm thinking that it might save me in case of bugs one might experience in 24/7 use in consumer drive firmware.

If you ask my opinion about this, I think this is the stupidest thing anyone can think of. I know it works, but what this basically means is that you broke the first rule when configuring any raid(use same brand and model). By doing that you basically put the entire RAID config at risk. The irony is that you increase the probability of having a failure by doing something which you think that improves this.
I know that it works. Sometimes you are facing a situation where you have to replace a broken drive in an old raid array and that replacement is either too expensive or it is not available anymore. So you sacrifice "same brand and model rule", but when you do it like this, in first place, I do not see any "benefits"...

 

[PliotronX]

I would overprovision by 30% and then boom you have what enterprise SSD's arrive as. But seriously, Railgun's suggestion is sound on the consideration of reads and writes. Roles like WSUS should be taken into account as well. If you can shift software and roles off of the SSD and just have the core server OS on the SSD, the benefits outweigh the risk.

 

[Chicken76]

If you ask my opinion about this, I think this is the stupidest thing anyone can think of. I know it works, but what this basically means is that you broke the first rule when configuring any raid(use same brand and model). By doing that you basically put the entire RAID config at risk. The irony is that you increase the probability of having a failure by doing something which you think that improves this.
I know that it works. Sometimes you are facing a situation where you have to replace a broken drive in an old raid array and that replacement is either too expensive or it is not available anymore. So you sacrifice "same brand and model rule", but when you do it like this, in first place, I do not see any "benefits"...

I'm really intrigued by your answer redzo. What are the advantages of running mirrors with identical drives? And what are the disadvantages?
I can only see as an advantage, the predictable performance of the array, you know how one drive performs and you can guess pretty accurately how a certain array will behave.
As disadvantages, I see several:

• you have difficulty maintaining the "same brand and model" rule as time goes by
• identical drives used in identical manner will likely fail within short intervals
• bugs in firmware will surface at exactly the same time, potentially bringing the whole array down

 

[dave_the_nerd]

2-5GB of writes per day is trivial. Well within reason for a consumer level SSD.

Of course back up, back up, RAID is not a backup, and back up. But that applies to servers, period, not just ones using consumer drives.

As far as "identical drives in RAID" goes, it mostly applies to hardware RAID controllers (where the unpredictable/inconsistent performance between drives can throw the controller out of whack.) For software RAID, it doesn't matter as much, although:

1) slower drives will handicap faster ones.
2) smaller drives will cap the stripe size and waste space on larger drives.
3) if you're buying based on price/capacity, it will probably be cheaper to buy two of the best deal than one of them and one that isn't.

I wouldn't worry about the short interval failure thing with identically used same-model drives - it's too random, and chances are you'll either have a random cradle death failure that won't bother the other drives, or you'll replace the drives before they fail - clusters of wear-and-tear failures really only happen in large drive populations.

Firmware bugs aren't common or destructive enough to scare me either. At least not since the Great Sandforce Open Beta Test of 2011.

Especially if you have backups.

 

[dave_the_nerd]

FWIW, lower-end "enterprise" SSDs are basically the same thing MLC SSDs consumers buy, with more overprovisioning. (Well, and encyrption, power loss protection, and longer warranties.)

They're usually used as a fast read-only tier.

Used in conjunction with a smaller SLC drive for write-cache... hubba.

 

[ultimatebob]

The Tech Report ran a torture test on a bunch of consumer grade SSD's, and found that they can handle about 700 TB (Yes, TB) of data writes before they start to fail. Two of the original six they started testing with are still running after 2 Petabytes of writes. That's more data than most file servers would process in 20 years.

So, yeah, I'd feel safe putting them in low to mid range server. Anything higher end than that should probably be connected to a SAN, anyway.

 

[dave_the_nerd]

So, yeah, I'd feel safe putting them in low to mid range server. Anything higher end than that should probably be connected to a SAN, anyway.

Ah, but then should your SAN use SSDs? :sneaky:

 

[CoPhotoGuy]

Not to run a SQL server I wouldn't.

 

[Chicken76]

Not to run a SQL server I wouldn't.

Why not?

 

[Carson Dyle]

Why not?

Probably for the same reason that his mom put vinyl slipcovers on all their living room furniture.

 

[ultimatebob]

Ah, but then should your SAN use SSDs? :sneaky:

A lot of them use SSD's for read caching, so yes.

 

[PliotronX]

Why not?

Because you touch yourself at night.

 

[SSBrain]

The main argument against the usage of consumer SSDs in servers is the manufacturers' limited warranty. On paper, it's valid as long as the drives are used for their intended purposes (office/home user scenario) and their host writes total doesn't exceed manufacturer specifications, which for consumer SSDs are usually exceedingly conservative on that regard (as a sidenote, Kingston is pretty much the only mainstream SSD manufacturer with honest endurance ratings for its consumer line-up).

This means that if a drive used that way fails during the warranty period, even if it's clearly not due to NAND wear, you might not be able to get a replacement/refund.

 

[Cerb]

Not to run a SQL server I wouldn't.

I would, gladly. An SQL server is not special. Really, it's not. If you don't have enough writing going on to need the extra endurance, and aren't running a SAN, or high-latency NAS, it's all a matter of who warranties it and how (with a SAN, especially, you'll really want to make sure anything you gave an ack back to gets written, to the best of your ability, without slow as death sync writing; and the Intel 730 is the only "consumer" drive still out there that can do that, and not by original intent).