FLASH/SSD: Ready for Primary Storage?

By Ashar Baig 1

Solid state drives (SSDs) have much to offer the data center—they’re faster, generally more reliable and less expensive to buy and maintain than hard disk drives (HDDs). But are they ready for data center prime time as primary storage? Let’s take a look.

SSDs offer faster data access—as much as 1,000 times the random input/output performance of hard disk drives (HDDs)—and greater data center energy efficiency—both the size of the drive and the amount of energy used in the drive is much lower. Plus, the input/output operations per second (IOPS) on SSDs is lower because the SSDs can perform much higher interactions per second, and have a much lower latency compared to HDDs.

Technology Comparison – HDDs vs. SSDs

HDDs magnetize parts of the disk and store data sequentially. Historically, HDDs have increased in capacity but not much in speed.

Built from NAND flash memory, SSDs store data persistently but not sequentially. They have no moving parts, utilize memory chips rather than spinning disk, and store data utilizing an algorithm sitting on the chip controller. Each SSD has eight, 16 or 32 chips, with each chip holding 4GB, 8GB, 16GB, 32GB or (soon) 64GB in capacity.

Data is written in blocks (think of them as pages) of 512KB or 2MB. When a block is flushed, all data is gone and is not recoverable. Even for a small change to the block, the document has to be written to a new block and the older version is slated for deletion.

Types of SSDs

Endurance

Flash has exceptionally fast write speed, yet it is prone to much high wear rate than magnetic drives. SSDs eventually wear out with use, due to the nature of flash media. Program erase (P/E) cycles, in particular, slowly wear out flash media. Reads, on the other hand, have no effect on endurance and can be sustained indefinitely.

What’s more, the oxides in flash wear out much faster over time, limiting the number of writes over the life of flash. It is like paper being erased: One can only erase the paper so many times before it starts to tear off.

To put it in context: In the flash world, if the read costs you a penny, the write will cost you a quarter.
Cost per terabyte-written is often the yardstick to measure and evaluate the various offerings in the marketplace. When you take into account the endurance of a solution, one that initially seems less costly over time actually may be more expensive.

Flash endurance is measured in P/E cycles, and its memory cells can endure anywhere from 3,000 to more than 100,000 cycles. Flash’s cell architecture is the reason for the variance.

SDDs suffer from degraded performance over time—they actually slow down with use over the life of the drive. That’s because of certain issues with the flash memory “garbage collection” and other things that are going on in the background.

SSD warranty specs spell out the amount of total writes the drive can be warranted for. SSD manufacturers often use a 50/50 mix of random and sequential writes to specify endurance, which is not a real-world scenario. Some will write random blocks of data but do not reveal the block size, and oftentimes the workload is not published, making it difficult to conduct an apples-to-apples comparison of the manufacturer-specified endurance rating. To put the SSD endurance claims to the test, the worst-case workload test is using 100 percent random writes of 4K blocks.

Total Cost of Ownership

SSDs drive greater data center energy efficiency by reducing the overall data center space and energy requirements. In the long run, SSDs can be cheaper on an IOPS basis.

Because of their high initial cost, SSDs are slow in replacing HDDs in primary storage environments for mission-critical workloads. Today, SSDs are used primarily in hybrid environments where flash is used for warm or hot data workloads. Additionally, flash is used for transactional workloads and for niche applications such as virtual desktop infrastructure (VDI)—80 percent writes and 20 percent reads, bursty and unpredictable, highly random I/O streams, very high-performance requirements—as well as within PCIe cards in servers, flash arrays and flash appliances.

In real-world IT environments, only a handful of apps can be classified as mission-critical. One should look for apps that can specifically benefit from the higher performance of flash. (Some may argue, however, that one never knows which apps will be hot from day-to-day.) Some flash-friendly apps include virtual environments, PostgreSQL databases, transactional data and high-performance computing.

Ashar Baig is president and principal analyst and consultant at Analyst Connection, an analyst firm focused on cloud computing, IT products and services and managed service providers. He has more than 18 years of high-tech industry experience. Baig also is founder and manager of the LinkedIn Cloud Backup group.