Balancing Performance, Capacity, Endurance in Enterprise SSDs

Posted on July 19, 2016 By Christine Taylor

RssImageAltText

Flash is now widely accepted in the data center. Many new storage systems have either flash tiers or all flash. Even newer HDD systems usually have the option to add flash at a later date.

Although the SSD industry sells more units to the consumer market, flash makers report that enterprise SSD sales are up by more than 65 percent this year compared to 2015. Samsung has the lion’s share of the 2016 SSD market and SanDisk, Toshiba, Kingston, and others all do good business selling consumer and enterprise SSDs.

The majority of flash products in this environment are enterprise-level NAND flash SSDs. The data center is not restricted to this architecture by any means. DRAM, for example, is highly suitable for high performance computing (HPC) environments, and server-side flash can accelerate high IO/low latency applications better than can most flash storage systems. But these options are not in general usage given DRAM’s high price and volatile memory and server-side flash’s restriction to single servers.

Thanks to its lower price and non-volatile memory, NAND flash SSDs comprise the majority of flash storage systems. (Keep in mind that NAND flash and SSDs are not synonymous. NAND underlies several different flash architectures including enterprise SSDs.)

SLC or MLC: How Much Does It Matter for Enterprise SSDs?

Enterprise SSD flash is usually built on Single Level Cells (SLC) or Multi-Level Cells (MLC). Storage buyers don’t usually get into the SLC/MLC comparisons—although perhaps they should. Vendors rarely present that level of detail, preferring to discuss price/performance/endurance numbers. However, flash cell types, along with the flash storage processor, are fundamental to that discussion.

MLC is found more often in consumer flash products such as cameras and phones. It’s less expensive than SLC as you would expect for a consumer play, but for general purpose storage, it can serve very well in an enterprise SSD. MLC performance can improve considerably when paired with a faster flash storage processor, which is why manufacturers like Intel spend development dollars on storage controllers for flash tiers.

The primary difference between the two architectures is that SLC stores one bit of data per cell, while MLC can read different states from multiple signals in a single cell. The MLC advantage is far more density, so the same amount of money spent on an MLC buys considerably more capacity than the SLC. The disadvantage of the MLC is that it has lower transfer speeds than SLC, and the higher cell activity translates to faster wear. An MLC cell might reach end-of-life after 10,000 erase and write cycles, while an SLC cell might last several times as long before failing.

In practice, this may not make a great difference since all SSD vendors use wear leveling technologies in order to spread wear around multiple cells. Additional features that guard against wear-driven failure are optimizing write operations so they write in chunks, de-duplicating writes, and preserving unused cells to substitute for worn cells. Nevertheless, in the enterprise storage space, SLC transfer speeds result in higher performance. In environments where budget and capacity are the larger considerations, MLC will offer greater density and thus lower cost-to-capacity.

Customers are not necessarily stuck with one choice. Flash array vendors often use both types within the same array: SLC for Tier 0 high-performance tiers and flash caches, and MLC for secondary flash tiers. This best-of-both-worlds alternative provides very fast performance for incoming data sets and preserves flash performance with larger data set capacity during the first few days or weeks of active data life. After that, the flash tiers will commonly tier to a nearline disk system or to disk tiers within the same array.

Key Enterprise SSD Characteristics

Let’s take a look at three core elements of SSD choices: performance, capacity and reliability/endurance.

  • Performance. Performance is SSDs’ main claim to fame over HDDs. SSDs have no physical heads to move around a platter, so communications between the SSDs and the storage controller is considerably faster. The speed of the storage controller and the presence of storage intelligence also impact a flash system’s SSD read and write performance.
  • Capacity. We already discussed the capacity/price difference between MLC and SLC. One of the driving factors is the percentage difference between your application’s reads and writes. If an application generates a high number of reads and a smaller number of writes, then a high-capacity MLC would do well. In this situation, capacity is more important than write performance. But if an application writes more than it reads, or if it’s a rough balance but performance is important, then the higher-priced SLC-based SSDs will serve data considerably faster with lower latency.
  • Reliability/Endurance. Bit Error Rate (BER) is important to know. BER calculates how many bit errors are occurring in NAND flash that are not subject to Error Correction Code (ECC). This does not necessarily mean that the SSD is ignoring the bit errors, only that the flash storage processor might be correcting it… or not, which results in the Uncorrectable Bit Error Ratio (UBER). Ultimately, no NAND flash SSD will be 100 percent error-free. Each write/erase cycle will affect cell integrity, and eventually the SSD will fail. However, any enterprise SSD should measure lifetimes in least three to five years thanks to low BER, wear leveling, and empty redundant cells that automatically replace worn cells.

Not every application needs the highest possible level of storage performance, but many of them need the speed boost that only flash can provide. Even a latency measured in milliseconds can delay an application for several seconds, which can result in frustrated employees and lost customer transactions. You need the speed of flash, and with falling prices, flash storage is available to nearly every company.

Having said that, there are SSDs, and then there are SSDs. Understand your data center’s needs: general purpose file-sharing can use lower-price, high-capacity flash storage while HPC will need the fastest possible performance from their SSDs. Know what you’re looking at when you’re in the market for a flash array.

Photo courtesy of Shutterstock.


Comment and Contribute
(Maximum characters: 1200). You have
characters left.