0

We all know SSDs are fast... but aren't all that great when it comes to any computer that writes to disk many times in any given minute.

Are there any SSDs that are suited for an enterprise environment? What am I looking for in an SSD that signifies that "This one is better for a work-horse server"?

Or maybe there isn't such a thing?

2
  • 1
    There are plenty of flash vendors dedicated to the enterprise market - I'm not going to name names because I'm biased, but depending on your requirements there are tons of options - all-flash SAN arrays, disks of normal form factors that you can use with a traditional RAID controller, and PCI-E cards with obscene throughput. None are cheap. Dec 29, 2011 at 1:51
  • Indeed, many are so non-cheap that you have to talk to a salesperson to get pricing.
    – Skyhawk
    Dec 29, 2011 at 2:35

5 Answers 5

4

Enterprise SSDs are rated for higher performance and endurance than consumer products. This is reflected in cost (I've paid $3000+ per SSD). Other answers cover the SLC (write-optimized) and MLC (read-optimized) differences. Look to the offerings from Pliant/Sandisk and STEC for options. These are the OEM products used by Hewlett Packard, Dell, NetApp, IBM, EMC and other storage hardware manufacturers.

3

There are 2 main types of SSD's in the market - MLC and SLC. MLC (Multi-Level-Cell) would be more of a 'consumer' version. It's less expensive than the SLC (Single-Level-Cell) version (considerably less expensive - by about an order of magnitude) but also has a significantly shorter lifespan. The SLC types are generally Enterprise oriented - they can handle 5-10 times as many read/write cycles as an MLC type SSD before a cell fails.

HOWEVER - having said all of that, the real question is: what are you planning on running on the SSD's? Even Enterprise solutions will use the MLC SSD's to save some money. By Microsoft's own calculation, in a heavy-hitting SQL environment (and we all know how disk-intensive a SQL Database can be), a 4-disk MLC SSD RAID would last 7-10 years before it became too degraded to use anymore. (Keep in mind, SSD's are designed with extra cells to replace the failed ones - this way you get maximum listed capacity until a critical threshold of cells fail, at which time you replace the SSD). An SLC SSD can last for decades before becoming unusable, and these are generally used in critical servers that are expected to be running for a very long time with very little maintenance.

Standard application servers rarely come close to the disk performance requirements of a DB server, so you'll need to clarify 'workhorse.' But generally speaking, if you build a server and use the branded SSD's offered (whether this is HP, Dell, IBM, etc...), the performance will be fantastic and the SSD's can be reasonably expected to last the lifetime of the server. Also, if they are in a RAID configuration, they can be replaced should one fail ahead of schedule. Intel, Crucial, and several other major vendors also offer solid SSD options. If you go that route though, be sure to check on the controller - some older/cheaper SSD's include older, poorer performing integrated controllers.

A final note: today, most RAID controllers max out at a throughput of about 150,000 IOPS. This equates to 4 SSD's, which is why you generally don't put more than 4 SSD's in a RAID group - after that, you're wasting performance - the controller can't move data any faster. However, I know that at least one major vendor will soon be offering controllers with better throughput - 300,000 and 600,000 IOPS - taking you to 8- and 16-disk SSD RAID groups for increased performance. I'd recommend you check with your vendor for this spec when you are configuring your server.

1
  • With your 7 - 10 year or decades lifetime estimate, how would enterprise drives last 7-20 times longer with significantly higher write loads than consumer drives? codinghorror.com/blog/2011/05/…
    – TheLQ
    Dec 29, 2011 at 18:29
2

We all know SSDs are fast...

Valid assumption

but aren't all that great when it comes to any computer that writes to disk many times in any given minute.

Say what? Sure, if you took a consumer-grade SSD and used it for log storage on a heavily loaded, highly transactional database server, sure, you're going to run out of write cycles very fast. That's the reason behind the huge cost difference between consumer drives and enterprise drives. With enterprise SSDs, you could pretty much keep them pegged at their IO limit constantly for years and not run out of write cycles.

This is a broad generalization, but enterprise SSDs will cost at least an order of magnitude more than their consumer-grade cousins.

Additionally, you will likely be purchasing two of them, configured in a RAID 1 pair, so keep that in mind when budgeting.

2

My understanding of SSD benchmarks is that writes of any kind are not challenging. The only challenge to a new SSD drive is mixed read/writes (80/20 to 20/80 mixed is the area of poorer performance). Random and sequential writes are harder on a SSD if it is nearing full as is the mixed read/writes.

Enterprise SSDs use Single Level Cell (SLC NAND) memory. Consumer level drives use Multi Level Cell (MLC NAND) memory. SLCs are more durable and do have a slight performance advantage.

A quick search for "Enterprise SSD" reveals that indeed there are plenty of products out there that are advertised to be more appropriate for long-term "enterprise use." However, for the price of a few SLC SSDs, you could get a stack of MLC SSDs at newegg and reap virtually the same speed benefits and simply replace the drives as they wear out.

I'd recommend searching Brent Ozar's blog for the term "SSD" and reading all of what comes back. It was through him that I considered the technique of using consumer drives en mass rather than "enterprise drives."

Hitachi, IBM, SeaGate and others all make SSDs that are marketed to the enterprise market. The trick is to see through the marketing hype. From my understanding of SSDs, just discern what level of IOPs you are dealing with and buy one that is rated to that level. Choose SLC if you have money to burn, or an array of MLCs if that can be fit into your scenario.

2
  • 2
    Note that if you are going to try the "simply replace the drives as they wear out" approach, you would be wise to have a strategy in place for monitoring wear. Some MLC drives, including the Intel 320 series, can report the percentage of wear endurance remaining as a SMART attribute.
    – Skyhawk
    Dec 29, 2011 at 2:31
  • @MilesErickson Yes, that would be advisable. =)
    – Wesley
    Dec 29, 2011 at 2:32
0

As much as I pretty much agree spot on with WesleyDavid, one thing to be aware of - many mechanical drives often slowly die - they often slow down terribly due to retrying bad sectors or start making funny noises, etc. Often when SSD's go, it will be a sudden and total catastrophic failure leading to loss of all data on the drive.

If you were unlucky enough for multiple drives in a RAID to go in a short period the RAID and data would likely be toast.

Although this where decent backups would save the day - whether to consider consumer grade / MLC technology depends on how high the cost of downtime to you is, and whether it would be the end of the world to have to go back to the last backup in the event of a problem.

2
  • Yes, mechanical drives do typically fail gradually. However, in an enterprise environment, you're going to be running them behind a RAID controller which will detect the UREs immediately, and will flag the drive as failed before it actually fails catastrophically.
    – EEAA
    Dec 29, 2011 at 3:37
  • Good point - although that would depend on having chosen a proper RAID controller from the start - e.g. I know from experience the Dell PERC 5i/6i would behave as you described, although the CERC would just keep retrying. - I believe the PERC is now sold as PERC Hxxx while the CERC is now called PERC Sxxx.
    – Robin Gill
    Dec 29, 2011 at 8:51

Not the answer you're looking for? Browse other questions tagged .