Tech2Tech
Applied Solutions 1
Solid state storage 101
Technology developments help support the enterprise with better performance, price and capabilities.
by Mark Nossokoff
Much is being said and written about solid state storage, especially solid state drives (SSDs). Some of the discussion is pretty cryptic and confusing. The bottom line is that SSDs are nearly ready for enterprise integration and can surpass traditional hard disk drives (HDDs) in capabilities and performance.
The basics
“Solid state storage” is a broad term used to refer to a class of storage other than rotating magnetic media, or HDDs. It is based strictly on semiconductor components and contains no rotating or mechanical parts. Several types exist, with dynamic random access memory (DRAM) and flash among the most prevalent.
SSDs are a specific type of solid state storage. These flash-based storage devices are packaged similarly to HDDs in form factor modules that are typically 3.5 inches or 2.5 inches for enterprise and 1.8 inches for consumer drives. They communicate with an I/O controller, host adapter or storage controller via a standard I/O interface such as Fibre Channel (FC), Serial Attached SCSI (SAS) or Serial Advanced Technology Attachment (SATA). Other packaging options for flash-based solid state storage include custom/proprietary form factor modules, standard PCI Express form factor cards and 19-inch rack mount appliances.
The following is a comparison of two types of flash technology devices that are integrated into SSDs:
- Single-level cell (SLC) is the more prevalent and mature flash technology, especially for enterprise applications. It provides higher performance and has a longer life but is more expensive and typically has less capacity.
- Multi-level cell (MLC) costs less and has less demanding requirements, making it the primary SSD choice for consumer and laptop applications. Vendors are working to improve the life and reliability of MLC flash devices so they are suitable for enterprise environments as well.
It’s likely that both types of flash memory technology will co-exist for a while, allowing solution providers to deliver multiple price/performance products for customers.
Flash life
A comparison of the devices’ program/erase (P/E) cycles and subsequent write endurance specifications shows why SLC has a longer life than MLC. (See table.) Each location on a flash device can be written to—or programmed—only a limited number of times. In current semiconductor processes, SLC can support roughly 100,000 P/E cycles, while MLC can support roughly 10,000 P/E cycles. Per day, this equates to writing the entire raw flash capacity 55 times for SLC and 5.5 times for MLC, assuming a five-year life span of each device.
SSDs employ a technique known as wear-leveling to ensure even and uniform P/E cycle distribution across the capacity of the device. If the device has usable and reliable capacity, the wear-leveling algorithm ensures that no hot spots are being written to on a flash device, as this overuse on a single location would cause the device to prematurely wear out.
Wear-leveling algorithms will evolve to further ensure uniform distribution of writes, as well as to improve the effective P/E cycle specification of the MLC, bringing it closer to the SLC P/E cycle metric. Several SSD suppliers are developing mechanisms to extend drive life, allowing MLC SSDs to be used in most enterprise environments.
As semiconductor processes continue to shrink, so will the number of P/E cycles a flash device can support. While algorithmic improvements will continue to allow flash for the near future, the time is coming when flash will have to be replaced by a new solid state medium within solid state storage devices.
Both flash technology and device design have significant effects on wear life. To determine the device most suitable for your organization, research each vendor’s specifications to determine whether a particular device has the lifetime required for a specific application.
Capacity
A flash device has both raw and user capacity specifications. Raw capacity is the entire amount of flash storage on a device. User capacity is the amount of storage available for user data. Wear-leveling algorithms, flash management that improves performance during writes and data protection algorithms all use some amount of capacity.
While it’s not quite a linear relationship, SLC and MLC support different capacity points. As implied in its name, an SLC flash device contains a single bit of information in each cell. Likewise, each cell in an MLC flash device contains multiple bits of information. SSD vendors typically offer MLC products that have twice the raw capacity of SLC products within a given semiconductor process.
Enterprise SLC SSDs range in capacity from 32GB to 256GB across the vendor’s offerings and are expected to roughly double every 12 to 18 months. MLCs are starting to emerge at the expected capacity points of at least twice that of SLCs and are expected to grow even faster as MLC technology evolves, allowing up to four bits per cell.
Performance
At the highest level, flash storage is much faster than HDDs but slower than DRAM. But performance across SSDs is much more difficult to compare.
Tremendous performance variability exists among solid state storage devices. Some of these devices are optimized for reads, while others are optimized for writes. On SSDs, most read performance is better than write performance. Also, all SSDs show varying degrees of uniformity (or lack thereof) and inconsistency in mixed read/write environments.
A clear understanding of the storage application’s workload, as well as knowledge of a specific flash-based storage solution’s capabilities, is critical in identifying the appropriate solution for your organization.
Flash-based performance is also a factor if the location on the device has previously been written to. Writing to or programming a specific location on a flash device is faster the first time it is accessed than it is to a location that has been written to and erased.
Ideally, standards will evolve so performance and other characteristics can be compared uniformly, reliably and accurately across different vendors’ products for both flash devices and storage systems. In the meantime, it’s vital that the application’s I/O profile be properly matched with the storage system’s capabilities.
Power
Solid storage devices are extremely energy-efficient, with power consumption typically ranging from 1W to 7W. This compares with roughly 10W for capacity-oriented HDDs (7,200 RPM) and 18W for high-performance HDDs (15,000 RPM).
At a system level, using SSDs in place of HDDs can reduce the power budget associated with physical storage by between 50% and 80%.
Cost
On a straight cost-per-gigabyte comparison, SSDs cost more than HDDs. In 2010, depending on the HDD vendor and capacity point, an SSD is expected to cost roughly 20 times as much as an enterprise HDD. Additionally, SLC technology costs more than MLC technology.
But cost per gigabyte should not be the only cost consideration. Depending on the application and deployment model of the solid state storage within the system architecture, one SSD could replace multiple HDDs to achieve the required performance. Certain applications are seeing or predicting a 12:1 or higher device reduction when replacing HDDs with SSDs.
If performance is the primary consideration, then cost per IOP should also be a key metric and comparison point. In this case, it is feasible that an SSD or some other type of solid state storage device is a more cost-effective solution.
The table summarizes the flash technologies and compares them with a high-performance disk drive’s characteristics.
Barriers to adoption
Even with the advantages that solid state storage has over traditional HDDs, as with any new technology, you should exercise some caution before diving headlong into broad, mainstream deployment. Take these issues into consideration:
Click to enlarge
Long-term reliability
With so few SSDs having been installed in enterprise-class systems, long-term reliability has yet to be fully proven, although early results are very encouraging. The write-endurance characteristic could become problematic if improperly designed SSDs are deployed in the wrong environments.
Vendors
The majority of SSD vendors generally fall into two camps:
- Flash experts with little or no enterprise storage experience or expertise
- Enterprise storage experts with little or no flash experience or expertise
Both camps have relatively difficult learning curves to overcome, with bumps and bruises along the way. It will be incumbent upon the storage systems providers to carefully and fully understand the origins of the SSD vendors when evaluating a solid state storage product.
Inflated performance expectations
Solid state storage provides significant performance advantages over traditional HDDs. However, if the overall system is not fully balanced and optimized to take advantage of the performance increase, then the overall results could be disappointing.
The same is true if the actual deployment doesn’t translate to substantial application performance improvements that are visible to the end user.
Interoperability and lack of standards
Given the differences in low-level algorithms and techniques utilized between the various SSD vendors (e.g., wear leveling, write conditioning, redundancy), it is difficult to compare traditional IOPs performance benchmark metrics among SSD vendors. Development and evolution of standards will eventually eliminate this confusion.
The right choice
Despite the minor challenges described, solid state storage­—particularly in the form of SSDs—is ready to be deployed today in specific enterprise storage environments such as data warehousing. Solid state storage performance benefits can be immediately achieved by matching it with the appropriate applications, thus avoiding its near-term limitations.
Mark Nossokoff, a senior member of LSI Corp.’s Strategic Planning team, is currently focused on solid state storage advancements for business, market and technology planning for storage systems.