Enterprise-class solid state drive (SSD) is most commonly based on NAND flash technology, though a number of systems are starting to feature DRAM chips
The difference between the two, in actual implementation, would not seem to be worthy of controversy. That alone makes it an ideal candidate for vendors to highlight differences other than price, though price will always be a factor in the purchase decision for the most expensive storage element in the data center. Where minor differences exist, one can expect FUD (fear, uncertainty and doubt) to play a role in some sales pitches.
One of the major advantages of NAND SSD is its nonvolatility, meaning that data is retained even without power. This is also true of NVRAM (where NV stands for nonvolatile). Conceptually, and perhaps architecturally, this matches well with the nonvolatile nature of hard disk drives. Certainly, it makes intuitive sense to have an entirely nonvolatile array to protect data in the event of sudden power failure. But is that fact or FUD?
DRAM is indeed volatile; take away the power and the data goes with it. However, data centers are protected by UPSes, plus arrays may have their own battery backup devices and many DRAM boards have their own on-board battery. One would need to be almost clinically paranoid to believe that data loss could occur. It may be hard to believe that data loss could occur.
It's axiomatic that SSD is great for read-oriented applications but suffers from a write performance deficit. DRAM, which is the primary technology of server memory, must operate with balanced read/write functionality. This would seem to favor DRAM over NAND, right?
First, let's clarify that "poor" write performance is still a relative measurement. Yes, compared to its read performance, NAND flash writes are substantially slower. Nevertheless, NAND write performance is still considerably faster than the writes of mechanical drives by as much as 10X. Moreover, I/O performance of any device depends highly upon myriad factors; vendor performance claims have enough caveats to choke a law firm. Finally, NAND vendors have invested considerable time and money in developing write-optimization strategies. Whether one technology is superior to the other may literally come down to specific application requirements.
NAND flash product life is not measured in time but rather in write operations. Erasing and writing to NAND causes it to gradually degrade and fail, but only at an individual cell level. Consumer-grade NAND SSD is rated for 3,000 to 10,000 write operations whereas enterprise-grade is rated for 100,000 write operations. NAND failure is usually gradual as individual cells fail and affect overall performance. Vendors sometimes over-subscribe their systems (i.e., include more memory than is actually claimed) to help compensate. And, as noted earlier, manufacturers have increasingly sophisticated write algorithms to minimize degradation. DRAM also suffers from hard errors, though time-tested methodologies are used to compensate or correct for them. DRAM doesn't have a write-life, but it certainly fails over time.
So, does DRAM have a longer life than NAND? Both offer product warranties lasting years. Many NAND vendors offer five-year warranties and some as long as 10 years. Some DRAM SSD warranties are similar. It's hard to give either a significant advantage here.
Weighing the claims
Although some vendors deploy distinctly DRAM or NAND devices, others are beginning to offer both. DRAM is more often positioned as a cache tier, in which case it is generally available as a performance layer. NAND complements DRAM by functioning in a hybrid storage pool where automated tiering software can dynamically move "hot" data to the fast SSD devices from slower HDDs.
DRAM versus NAND is much more likely to be a religious war to the vendors than it is to IT organizations. As such, most IT organizations will find themselves using DRAM and NAND rather than DRAM or NAND. Nontechnological factors such as architectural coherence, vendor preference and price will be much greater factors in purchase decisions. Even so, IT organizations will want to weigh the differences to optimize their DRAM and NAND deployments.
Phil Goodwin is a storage consultant and freelance writer.
This was first published in August 2012