RTC Magazine

Flash memory’s drastically dropping price in conjunction with its inherent non-volatility makes it a prime candidate for mass storage integration, although its write performance and endurance make such an implementation difficult. RAM on the other hand is extremely fast in both reads and writes, but its cost per capacity has slowed its enterprise adoption. These two storage technologies are now being merged together to create Cached Flash systems that leverage the capacity, price and non-volatility of flash with RAM’s unmatched speed and performance.

Solid-state disk, commonly referred to as SSD, uses a volatile (RAM) or non-volatile (flash) memory as the primary storage media. If the systems are built using RAM, there will always be battery backup to internal hard disk drives to preserve data even if external power is lost. From the point of view of the operating system, SSD systems attach and behave just like hard disk drives. It is important to maintain the distinction between RAM and flash memory, although both are “solid state” in nature, their capabilities and composition are far from similar.

Flash memory has been getting a lot of exposure and is a very appealing technology due to its inherent non-volatility, ruggedness, low power consumption and rapidly dropping price. This technology has been around since the late 1980s but was first brought to market in the early 1990s by Toshiba. Many of these factors make it a prime candidate to be a successful technology in the consumer market since it can be integrated into many existing technologies as an alternative for hard disk.

There are downsides to flash memory: its writes are extremely slow, coming in slower than hard disks in some cases, and it has limited write endurance, meaning that the chips will render themselves unusable in a matter of 10,000 to 100,000 writes depending on the type of flash chip used. Multi-layer chip (MLC) technology is specified to perform 10,000 writes before wearing out while single-layer chip (SLC) technology is rated at up to 100,000 writes.

Random Access Memory, or RAM, is a more mature technology than flash and can process both reads and writes at 10-15 nanoseconds, although it is inherently volatile and therefore requires a power source in order for it to operate and retain information on the chip. RAM has made quite a footprint in the mass storage arena as well, given its equal write and read performance and its capability of executing a seemingly unlimited number of transactions. An overview of the usefulness of different storage technologies for various data patterns is given in Table 1.

Consumer Applications

Flash is very attractive to the consumer market. Overall, the most appealing aspects of this technology are its ruggedness, density, low power consumption, price per solution and performance respectively. The most prevalent applications for flash are cameras, portable media and thumb drives. Recently both portable media devices and laptops have started integrating this non-volatile technology as well, but more for the purpose of hard disk replacement.

Thumb drives—also known as pen drives, USB drives and flash drives—are essentially small, portable, flash memory devices that connect to computers via a USB drive and can store information without a power source as well as quickly and easily enable sharing of information between computers. Ruggedness is the most important feature of this technology since it needs to survive everyday wear-and-tear activities such as being dropped or accidentally being put through the washing machine and still maintain its data integrity.

Density is important for this technology as well since, although this device is small, it is rather inconvenient to have to carry more than one around at a time, and if large file transfers are necessary, at least one gigabyte of data will be required to hold the desired amount of data. Low power is also essential, and although the amount of power needed to perform a transfer is not as important, the fact that the device can retain all the information with no power at all makes for a purely portable memory device. As flash densities improve, expect an ever growing list of applications for the USB thumb drive. It is not far-fetched to believe that most software and possibly movies could be distributed on this technology in the next five years.

Portable media devices, such as MP3 players and cell phones, have definitely been on the rise. With the innumerable variety of devices available and the public’s desire to have smaller devices with more capabilities, flash seems like the best way to satisfy those needs. Flash is the perfect fit for the same reasons that flash is the perfect memory choice for the thumb drive. It is rugged, dense and uses less power. The power feature is slightly more interesting in this integration given the fact that with lower power consumption, the battery life for the devices’ other functions will be much longer.

Using flash in laptops is the most recent integration of the technology. Flash drive manufacturers have had IDE and Serial ATA drives out for several years, but the recent drop in prices and increases in density are enabling the use of flash in the laptop. Large flash manufacturers can see that this segment is likely to grow rapidly, and every major player has a credible option for this category. Unlike thumb drives and portable media devices, the hierarchy of feature importance is a little different. Low power is the most import characteristic of flash that the laptop utilizes. This is because it will extend the battery life and therefore increase the amount of time in which the product is usable.

Density, of course, is still very important. The smaller memory footprint with the greater capacity will allow for more information and capabilities with less weight and overall size, a very appealing quality in today’s consumer market. Ruggedness is the next most important aspect of flash being integrated into laptops. Flash drives can survive g-forces that even the best-designed hard drive cannot survive, making the flash drive a data saving device for laptops that can be exposed to unexpected drops. Performance is also more important in this use of the technology than the other consumer electronics applications. The performance necessary for thumb drives and portable media devices is only just enough to provide for seamless data transfer and streaming media, and given flash drives’ fast reads, its speed will rarely, if ever, be insufficient. But use of flash in laptops can decrease the boot time of operating systems as well as increase the fluidity of applications ranging from typical multimedia, desktop applications to gaming.

Data Center Applications

Given flash memory’s slow write performance and poor write endurance, these systems have not been widely deployed for enterprise-level integrations. RAM-based solid-state disks on the other hand have been integrated into many enterprise-level applications. However, they are not as commonly found in the consumer sector since their price per gigabyte is rather significant, persisting data requires additional battery and backup storage, and consumers rarely have a use for such great performance. At the enterprise-level, RAM SSD is mostly used for latency-sensitive database applications such as: on-line transaction processing (OLTP), batch processing and data warehouse acceleration. These systems are capable of accelerating enterprise application performance up to 25x versus performance from hard disk drive-based storage.

RAM SSDs are physically added to datacenters via rack-mounted systems. They are capable of dramatically accelerating these applications by considerably lowering the response time of the storage. This translates into users completing processes faster. Therefore, the RAM SSD market focuses on applications where the value attached to completing a process faster—often for a large number of users—outweighs the RAM SSDs’ costs for the capacity required.

Writing to disk is a common cause of bottlenecks encountered within particular applications. There is a broad range of OLTP applications that require a write to disk before a process can complete—usually to a log file with a small capacity. OLTP’s dependence on storage performance combined with the large value attached to accelerating processes with an inadequate capacity for logging them is one of the reasons SSDs are heavily deployed in financial applications. Other industries have similar latency-sensitive OLTP processes that directly interface with the customer. Failing to respond to a search or purchase request can result in a dissatisfied customer and lost business.

Another latency-sensitive application is batch processing. Batch processes have set deadlines that need to be met in order to maintain operational efficiency and compliance as well as a variety of other reasons that are unique to the organization. In a database context, each process may result in millions of I/O operations and CPU cycles. The I/O component of the process can be accelerated significantly with RAM SSD by shortening the time for each run. This allows the business to utilize the components tied up with unassisted batch processing for other goals.

Price per capacity, although dropping for RAM SSD, is still more than many companies are willing to spend considering the capacity they require for their solution. So, where is the product that has the high performance of RAM SSD and the inexpensive, rugged, non-volatile capacities of flash memory? Until now, there was no such beast that would even be considered “enterprise-grade” flash storage. However, storage companies are starting to test flash hard drives in their RAID enclosures, and companies that have focused on RAM SSDs are looking for ways to take advantage of flash memory’s benefits while isolating its weaknesses. A new type of system has resulted from this called Cached Flash solid-state disk. These systems are a blend of large RAM cache (up to 64 Gbytes) and RAIDed Flash memory (Figure 1).

The best systems in this category are designed from the ground up to take advantage of the strengths of the respective technologies, however it can be expected that many designs will just involve replacing the hard disk drive in a storage array with a flash drive. DDR caching is used to isolate the application from the slower write performance of flash memory while simultaneously protecting the flash memory from small block random writes. DDR caching is used more for its write performance as read performance from the flash memory subsystem should be excellent. Large arrays of flash memory are possible due to the high densities of flash chips and the fact that the chips can be packed densely in their enclosures. Additionally, because flash does not require power for data persistence, these large arrays of flash memory use very low power. Cached Flash systems designed for the enterprise will also have many redundant layers of data protection to isolate the application from bit errors observed in the flash memory and even from errors observed in the DDR cache.

Cached Flash systems are able to offer performance levels that are not possible with single flash hard disk drives because of the massive parallelism possible in larger form-factors with higher chip counts. It is reasonable to expect one of these systems to sustain over 100,000 read operations per second from the flash memory and much higher rates if only the cache is accessed. Another feature of large cached flash arrays is the ability to support bandwidths as high as 2 Gbytes/s. Cached Flash systems will effectively fill the gap between capacity, performance and price that enterprise systems have been lacking since the inception of the data center.

Texas Memory Systems
Houston, TX.
(713) 266-3200.
[www.TexMemSys.com].