By Greg Schulz, Server and StorageIO @storageio

Solid-state devices (SSDs) are data storage memory (Figure 1) mediums that utilize semiconductor based memories as opposed to magnetic media found in hard disk drives (HDDs) or magnetic tape.  Semiconductor memories include ultra fast volatile dynamic random access memory (DRAM) commonly found as main memory (e.g. RAM) in servers along with and non-volatile memory (NVM) typically NAND flash. Nand flash based SSDs can be found in cameras (as SD cards), cell phones, iPods, and PDAs, as well as in notebooks, net books, laptops, tablets, and workstations. SSDs are also appearing in larger servers, appliances, and storage systems from consumer to enterprise level.

Figure 1: Memory and data storage map

Beyond the benefits of using flash-based storage for consumer products, SSDs are gaining popularity in cloud, virtual and traditional data centers to address I/O performance bottlenecks and improve storage effectiveness. A challenge for many environments is the decades old issue of addressing the server–storage I/O performance gap (Figure 2) which is discussed in more depth here (http://storageio.com/Reports/StorageIO_WP_080706_Cover.pdf ).

Figure 2: SSD enabling closing the storage IO performance to capacity gap

The cost of hardware continues to decrease, while servers are becoming faster and denser. Storage capacity and availability also continue to increase, while physical footprint and price decrease, but there is still a gap between storage capacity and server and storage performance. The result is that, for some applications, to achieve a given level of performance, more resources (disks, controllers, adapters, and processors) are needed, resulting in a surplus of storage capacity.

Nand flash based memories have risen in popularity given their low cost per capacity point and that no power is required to preserve the data on the medium. Flash memories have gained ground at the low end, ranging from SD cards or chips for cameras or smart phones to USB thumb drives and MP3 players, given their lower power, low cost, and good capacity points while preserving data when powered off. There are two common types of flash memories, single level cell (SLC) and multilevel cell (MLC). SLC flash stores a single data bit per memory cell, whereas MLC stacks multiple data bits per cell, enabling higher capacities at a lower cost. Typically, MLC based flash is found in solutions that have a higher capacity per price point. An example of a nand MLC flash die is shown in figure 3 (courtesy  of IMF: Intel Micron Foundry) representing an aggregate 128Gbits (16GBytes) using 20nm technology. Multiple nand flash dies (figure 3) are stacked in a chip package that are then combined with a controller and installed on a card such as a DIMM, PCIe card or in a common disk drive form factor.

Figure 3: Example of MLC nand flash 20nm die courtesy of IMF

Persistent vs. volatile memory storage

Semiconductor nand flash memories are non mechanical meaning no physically rotating or moving parts for accessing the magnetic media. Note that I said no mechanical or physically rotating parts as opposed to popular industry folklore that SSDs do not have moving parts. What this means is that with SSD the cells where data are stored do not have to be mechanically moved to where a read/write head and read or write the magnetic stored data. Instead low level controllers coordinate and access the columns of data stored in a semiconductor cell.

With DRAM the data is randomly accessed and exists as long as a charge or power is applied to the memory resulting in a very fast, however non persistent storage medium. With nand flash, being nonvolatile data stored in cells is not lost when a charge or power is removed being persistent similar to magnetically recorded data. The benefit of persistence memory (figure 4) is that data is retained when power is removed similar to magnetic data stored on a HDD or tape. Another benefit of nand flash is that by being semiconductor based, it is faster than HDDs and at a lower cost per capacity than faster DRAM.

Non-Persistent

Persistent

Dynamic Random Access Memory (DRAM)

Nand flash (SLC, EMLC, MLC), Phase Change Memory (PCM), NVRAM

Power required to store and retain data

No power required to retain data

Data lost when power turned off

Data retained when power removed

Battery backup used for data retention

No need for battery backup

Fast on both reads and writes

Reads faster than writes, writes are many times faster than those of a HDD. Some will be faster on writes vs. others with writer gather or other optimization.

Does not wear out due to use or cycles

Firmware and controller optimization to implement wear leveling maximizing duty cycle and endurance

More expensive than nand flash

Higher cost vs. HDD however lower cost than DRAM

Found in computer servers, cache in storage systems

Found in cameras, cell phones, PDA, MP3, servers and storage

Packaged as Dual Inline Memory Modules (DIMM)

Packaged in DIMMs, SD cards, thumb drives, drive form factors with SAS and SATA interface, PCIe cards, servers, storage systems and appliances.

Figure 4: Non-persistent (volatile) and persistent (non-volatile) memories

Does that mean nand flash does not have moving parts? Well, technically speaking there are no mechanically moving parts, and contrary to a myth that I have heard from some sales people SSDs do in fact consume electrical power thus generating heat. Nand flash memories are cell based that while not mechanical, there are in fact microscope electrical gates within the silicon dies that  form the cells where data are stored that can wear out over time. These cells or gates are what enable data to be retained when power is removed and have design cycle life of how many times they can be written to. DRAM cells on the other hand do not have duty or life-cycles per say such as nand flash, however keep in mind that when you remove power from DRAM you also lose your data. Also keep in mind that magnetic media, as is the case with any medium when properly cared for and used have a very long lifespan as well often being limited by their mechanical parts or other interface components.

Read more in part II of this series (The call to duty, SSD endurance)

Share this Article:

Share Button
Greg Schulz (17 Posts)

Greg Schulz is Founder and Sr. Analyst of independent IT advisory and consultancy firm Server and StorageIO (StorageIO). He has worked in IT at an electrical utility, financial services and transportation firms in roles ranging from business applications development to systems management and architecture planning. Greg has also worked for various vendors in addition to an analyst firm before forming StorageIO. Mr. Schulz is author of several books (Cloud and Virtual Data Storage Networking – CRC Press, The Green and Virtual Data Center – CRC Press, Resilient Storage Networks: Designing Flexible Scalable Data Infrastructures – Elsevier), active with social media with his engaging approach and a top ranked blogger. He has a degree in computer science and master’s degree in software engineering. Learn more at www.storageio.com

Connect with Greg Schulz:


Related Posts:

31 comments for “SSD options for Virtual (and Physical) Environments: Part I Spinning up to speed on SSD

Leave a Reply

Your email address will not be published. Required fields are marked *


8 + six =