3

I am in need of a terabyte or so of storage space for Lucene indices on a testing server. I'm having a hard time differentiating between disks like the Samsung 840 evo for about $450 vs SSDs with similar capacities that cost many thousands of dollars.

What concerns should I have regarding the 840 evo, and what should I be looking for? This server will be used strictly for in-house testing purposes by our development team and will have to put up with periodic load testing.

  • 2
    The big difference, which has yet to be spelled out explicitly and clearly is that the "many times more expensive ones" have a much higher write-erase cycle limit before wearing out. (How many times a cell can be written to before it becomes unusable.) For end-consumer uses, you don't really need more than the hundreds of thousands of write-erase cycles you get on consumer gear. For an enterprise, which could be using an SSD for a database that's written to multiple times a second, you just might need the extra write-erase durability the expensive ones offer. – HopelessN00b Jan 19 '15 at 20:58
  • What would you see as the major differences between storagereview.com/samsung_ssd_845dc_evo_review and neweggbusiness.com/Product/Product.aspx?Item=9B-20-147-247 for the described use case (internal testing, periodic load testing)? – balazs Jan 19 '15 at 21:31
  • Honestly, it doesn't sound like you'd get much value out of the more expensive SSDs. I'd go with the consumer-grade one, and if you managed to wear it out, I'd be surprised, but you could replace it with a new one and still end up saving thousands of dollars. – HopelessN00b Jan 19 '15 at 21:34
  • Where would I get value out of it, shy of operating a data center? Backblaze doesn't seem to suggest the enterprise drives are worth their price either; backblaze.com/blog/enterprise-drive-reliability – balazs Jan 19 '15 at 21:54
  • In applications with very high write loads. If you have to ask, you probably don't have one, and probably don't have any use for the more expensive SSDs. – HopelessN00b Jan 19 '15 at 21:58
2

Differences are:

  • Intended market/usage. Enterprise drives are meant to be on 24/7 and can usually withstand higher temperatures or other stresses. Consumer grade drives are expected to be on 8-10 hours a day or so. Lifespan and MTBF are based off these expectations.

  • Features included/supported by firmware. Enterprise drives may have more intelligent wear-leveling, better error-checking, and 'smarter' firmware in general. Enterprise drives may also have a larger capacity of spare overhead to deal with failed blocks. Consumer grade drives will have a stripped down or more basic firmware.

  • Component/construction. Enterprise drives will usually be made of longer-lasting flash memory, and will use different architectures that may be constructed to optimize performance in specific workloads. Consumer grade drives will be made of cheaper components.

  • Warranty and support. Enterprise drives have better warranties and better support, allowing customers things like next-day replacement and advance RMA. Consumer grade drives will usually only be replaced via an RMA process where you do not have a replacement for several weeks.

  • Hardware level support, QVLs. Enterprise grade drives will be tested & vetted, and will be guaranteed to work in particular platforms. Consumer grade drives will be expected to work in consumer grade computers, and are not vetted for other platforms.

In general, if: 1.) you are very comfortable with supporting your own hardware, and 2.) you have a good understanding of how to avoid data loss through redundancy/backups, then there is no real problem with using consumer grade gear.

  • 2
    Keep in mind that companies will only differentiate if it means real savings. "Worse firmware" means developing firmware twice, and that's more expensive. If the enterprise firmware works on consumer-line products, then ship it. Less warranty, OTOH, is a straight saving. – MSalters Jan 19 '15 at 20:44
  • 1
    There is one key reason going well beyond mere selling points like manufacturer warranty (which are only of concern as long as buying spares is a problem or is becoming too expensive) or brand: enterprise drives typically come with supercaps and conservative caching policies where you can be sure that data acknowledged as "written" has been permanently stored. See lkcl.net/reports/ssd_analysis.html for an interesting field study of SSD power outage reliability. – the-wabbit May 23 '15 at 9:10
  • Much of this answer is outdated. There are no "higher temp/stress" drives. they may have been tested at higher temps but they are the exact same components. Firmwares are now virtually the same, as they have the same goals, support is better, but there is no "longer lasting flash" and afaik, no one has found a component difference (other than higher voltage capacity in enterprise drives due to plane failure tolerances) In short SSDs are no more reliable than platters- and consumer makes 0 difference in reliability. tomshardware.com/reviews/ssd-reliability-failure-rate,2923.html – Jim B Mar 20 '16 at 2:38
0

The 840 EVO is not really made for professionals but more for normal users, which is why the price is lower. Many of the features that are very useful for servers are not present here, such as error-fixing technologies, High Endurance Technology (found on some Intel drives for datacenters), and of course performance. All these features don't necesarily justify the price, but companies have more money to spend than individuals, so by making them paying that much, manufacturers can cover their development fees and reduce the price of ther consumer-grade products. I don't think you even need an SSD at all for what you want to do. I'd personnaly go for 10,000 RPM hard drives to make a RAID, but it all depends on your budget.

  • 1
    Even HDD's have Error Correction Codes nowadays; the reliability would be abysmal otherwise. Even the cheapest USB flash is likely to have ECC. Simple reason: using ECC is actually cheaper, as you can cram more bits in that way. In particular, you can use MLC flash on designs that otherwise would only work as SLC. – MSalters Jan 19 '15 at 20:41
  • We've been using 840 Evo's on some servers in production. They have been running error free for over a year without problems... they are not heavily loaded servers though. – Matt Aug 9 '15 at 23:10
0

In the specific case of Samsung consumer vs Enterprise SSDs, the differences are:

1: Write endurance - The enterprise drives will handle more writes

2: Powerdown protection - If the juice goes off, they'll flush their buffers

3: Guaranteed write latency - many consumer grade drives end up pausing for a few tens or hundreds of milliseconds every so often whilst they do wear levelling. Samsung's enterprise drives don't do this.

4: Spare blocks - the enterprise drives have a lot more of these.

5: Better internal error correction - the industry standard 1 in 10^14 bits works out to one uncorrected byte emitted from a drive every 5-8Tb of reads and based on what I see in my disk arrays that's about right. Moving to 1 in 10^15 is ten times better etc.

All Samsung drives use MLC or TLC technology, even the enterprise ones.

Whilst SM or PM-series drives are are significantly more expensive than EVOs, bear in mind that EVO drives are designed for light duty write cycles and may suffer more from write amplification - experience in the consumer sphere shows that most people fill drives with static content (photos, etc) and leave them for long periods - a occasional bit error here or there in an image doesn't matter. It does in a spreadsheet (which is why you'd use PRO drives) or when crunching large quantities of scientific data (which is where I use the enterprise drives)

Comparing the consumer -PRO range with the enterprise range shows a 15-25% premium which for many uses is justifiable (unlike spinning media where an enterprise can be 3 to 4 times the price of an equivalent consumer drive and Backblaze is quite right in their assessment of mechanical drives)

I'd be happy to drop EVOs in a desktop or games machine, less so if I was using it for business purposes. If hammering the drives then enterprise models are the way to fly - but consider something like a ZFS array with SSD cache drives (eg: FreeNAS) as this would be much cheaper and suits many modelling processes where only a small segment of data is being worked on at a time.

No matter what, when you have these quantities of data, back them up. RAID only protects against some kinds of hardware fault, not accidental deletion or catastrophic events.

-1

The main difference is the Flash technology in use.

Server grade SSD use SLC technology (Single Level Cell) where each flash unit (cell) stores one single bit of data. More reliable, more expensive.

Consumer grade SSD use MLC technology (Multi Level Cell) where each unit stored up to 4 bits of data. Less reliable, far cheaper.

With MLC you can squeeze 4x more data to the same amount of silicon than with SLC and the SSD is therefore a lot cheaper. The trade-off is that MLC wears out a lot faster than SLC. However for most common usage scenarios MLC technology is perfectly fine. Even for "normal" servers like yours you should have no problems if you put the disks into some sort of RAID.

A great article about MLC vs SLC is here: http://www.computerweekly.com/feature/MLC-vs-SLC-Which-flash-SSD-is-right-for-you

  • Enterprise MLC drives exist. SLC is rarely a consideration in my deployments now. – ewwhite Jan 20 '15 at 5:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.