Demartek SSD Primer
Updated 19 April 2012
By Dennis Martin, Demartek President
Because of the great interest in solid state storage (SSS) technology, we have compiled this summary document providing some basic information for solid state drive (SSD) or solid state disk technology. The terms SSS and SSD are often used interchangeably. Technically, SSD refers to a specific form factor, while SSS refers to any solid state storage form factor.
Solid state storage devices are computer storage devices that use memory technology for the storage media rather than traditional magnetic media such as hard disk drives (HDD) or tape drives. These SSS devices can be made with DRAM memory technology or Flash memory technology, or sometimes both. These devices appear to the host operating system as storage devices.
This document will be updated periodically and may become larger over time. Contact us if you’d like to see additional information in this document.
The Demartek SSD Deployment Guide 2012-Q2 is now available. This guide covers NAND flash characteristics such as SLC, MLC, endurance and performance. It also discusses SSD form factors and data placement strategies such as caching and tiering. A similar deployment guide that we have already produced is the Demartek iSCSI Deployment Guide 2011.
More information
To be notified when we post new deployment guides, additional tips and reports, sign-up for our free monthly newsletter, Demartek Lab Notes. We do not give out, rent or sell our email list.
Contents
- Acronyms
- Form Factors
- NAND Flash Basics
- NAND Flash Endurance
- NAND Flash Performance
- Cost Metrics
- Data Placement: Caching vs. Primary Storage
Acronyms
- EFD — Enterprise Flash Drive
- SCM — Storage Class Memory
- SSD — Solid State Drive (or Disk)
- SSS — Solid State Storage
- SLC — Single-Level Cell
- MLC — Multi-level Cell
- eMLC — Enterprise MLC
- P-E Cycle — Program-Erase Cycle
- GB — Gigabyte
- MB — Megabyte
- PB — Petabyte
- TB — Terabyte
Form Factors
SSS technology is available in several form factors. These can be:- Disk drive — 3.5-inch, 2.5-inch, 1.8-inch, etc.
- PCI Express (PCIe) adapter card
- Memory slot — DIMM, SO-DIMM, etc.
- External accelerators for SAN storage or NAS devices
- USB drive
NAND Flash Basics
NAND flash is a type of flash memory that is similar in basic function to Electrically Erasable Programmable Read-Only Memory (EEPROM). The underlying technology is a floating-gate transistor.Data in NAND flash is erased and programmed (written) in blocks. This process is known as the program-erase cycle. NAND flash blocks are typically 4KB in size, though some are larger. NAND flash is low-power, low-heat, low-weight and low-noise when compared to hard disk drives.
NAND flash is available in two basic types: Single-level Cell (SLC) and Multi-level cell (MLC). SLC flash has one bit per cell, and is generally designed for enterprise applications. MLC flash has more than one bit per cell, and is generally designed for consumer applications. There is a relatively new category known as eMLC (Enterprise MLC) that is MLC flash but with some of the characteristics of SLC flash. The enterprise characteristics of eMLC flash are provided primarily by the intelligence within the low-level flash controllers on the device.
MLC flash is available as two bits per cell, three bits per cell or four bits per cell. These are generally known as MLC-2, MLC-3, and MLC-4. Three bits per cell is also sometimes known as Triple-level Cell (TLC). As more bits per cell are added, capacity increases while performance and endurance decreases. The chart below provides a sliding scale of various factors for each of the types of flash.
 |
SLC | MLC-2 | MLC-3 (TLC) |
MLC-4 |
|---|---|---|---|---|
| Bits per Cell | 1 | 2 | 3 | 4 |
| Performance | Fastest | Slowest | ||
| Endurance | Longest | Shortest | ||
| Capacity | Smallest | Largest | ||
| Error Probability | Lowest | Highest | ||
| Price per GB | Highest | Lowest | ||
| Applications | Enterprise | Mostly Consumer | Consumer | Consumer |
SLC flash typically has 10x - 20x better endurance than MLC-2 flash. SLC has a typical life of at least 100,000 write (program-erase) cycles per bit. MLC-2 flash has better endurance than MLC-3 or MLC-4. MLC-2 flash has a typical life of 3000-10000 write cycles per bit. MLC-3 flash has a typical life of 300-3000 write cycles per bit. Enterprise MLC (eMLC) generally uses MLC-2 flash and has advanced features in the flash controllers that are able to typically provide 20,000-30,000 write cycles per bit.
NAND Flash Endurance
There has been enough concern about NAND flash endurance, especially for enterprise applications, that the memory standards body, JEDEC, has defined standards for NAND flash endurance. They have divided NAND flash devices into the following two categories:- Client — 8 hours per day of active use
- Enterprise — 24 hours per day of active use
NAND Flash Performance
The storage standards body, SNIA, has created performance testing standards for SSS technology. SNIA has defined performance testing standards for two classes of NAND flash devices:- Client — Deployed in single user desktop or laptop systems used in home or office environments
- Enterprise — Deployed in servers in data centers, storage arrays and enterprise-wide multi-user environments
Demartek conducts performance testing for SSD devices and uses the SNIA SSS PTS or a subset of it, depending on the specific type of testing required.
Cost Metrics
Many people are familiar with cost per gigabyte ($/GB) for storage pricing, and have attempted to compare SSD pricing and HDD pricing on this basis. While this is one reasonable measure, there are other metrics that are appropriate for SSS technology. The following table provides general ranges of these other metrics for the device types as a whole. Individual devices will have specific metrics that generally fall within the ranges described below. The metrics for eMLC devices tend to between the SLC and MLC metrics shown below, with some overlap. |
$/GB | $/IOPS | IOPS/Watt |
|---|---|---|---|
| SSD (SLC) | $10-$40 | $0.005-$0.15 | 1000-15000 |
| SSD (MLC) | $1-$3 | $0.004-$0.05 | 1000-15000 |
| HDD (Enterprise) | $0.50-$1.00 | $1-$3 | 10-30 |
| HDD (Desktop) | $0.05-$0.10 | $1-$4 | 10-40 |
Our conclusion based on this data:
- SSDs are dollars per gigabyte and pennies per IOPS
- HDDs are pennies per gigabyte and dollars per IOPS
Data Placement: Caching vs. Primary Storage
A key factor for SSD technology is deciding how to place data on these devices. SSD technology
can be deployed for data caching, primary data storage, or both. Solutions for both types of data
placement deployments are available from individual devices in a server up to large-scale, datacenter-class
storage arrays. We have conducted some performance tests using these two data placement techniques.
These evaluation reports are available in the Demartek SSD
Zone. Here is an overview of these two data placement strategies and techniques.
Caching
- Caching controller identifies any frequently accessed data (“hot data”)
- Caching controller automatically moves a copy of the hot data to SSD media
- Multiple applications can benefit from the SSD cache simultaneously
- Performance improves over time, as cache is populated with data
- Overall remaining HDD I/O load is reduced: fewer I/Os go to the HDDs
- Some caching solutions only cache reads, others cache both reads and writes
- Caching is relatively simple to manage
Primary Storage
- User decides what data to place on SSD
- User decides when to place data on SSD
- User moves specific data to SSD
- SSD benefits only the applications that use the data placed on the SSD
- Performance improves instantly
- Automated tiering software can help select and move data to SSD, based on policies
Some vendors who provide only one of these solutions will probably provide the other type of solution in the future. Some vendors allow SSDs to be split into two parts, one dedicated to caching and the other dedicated to primary storage possibly with their automated storage tiering software.