Caching in on Solid-state Storage
Intelligent use remains the best way to exploit speed and maintain the highest possible ROI
Rob Farber
Solid-state storage is revolutionizing computer storage. Unlike the currently ubiquitous rotating media disk drives, solid-state disk drives have no moving parts to waste power or delay data accesses. The fastest PCIe-based solid-state devices can perform over a million random disk accesses per second, while the fastest rotating media disk drives can deliver around 200 random disk accesses per second.
Solid-state disks are part of the cachet of “cool” laptops that elicit oohs and ahhhs from admiring users both for speed and low power utilization. These same benefits translate to enterprise computing, although the admiration comes from wonderful return on investment (ROI) and power savings. Price is the reason why 5,000 times slower rotating media disk drives remain popular, but multi-level cell (MLC) solid-state disk (SSD) continues to erode the price difference. The intelligent use of solid-state storage through the use of caching or tiered storage systems still remains the best way to exploit the speed of solid-state storage and realize the best price/performance ratio.
There are two principle types of solid-state technology on the market at this time: MLC and single-level cell (SLC) flash drives. MLC-based storage has an increased storage density, because it stores bits of information in multiple levels of a cell. This translates to lower manufacturing costs that can be passed on to the consumer, but it also translates to a higher error rate. Most consumer SSD products are based on MLC memory. Various error correction schemes make it extremely unlikely that a consumer will ever see an error from their solid-state storage device.
SLC technology stores bits of information on a single level per cell. This decreases power consumption even further and increases transfer speeds over MLC memory. SLC memory is primarily sold to enterprise customers, because the advantages of a lower error rate coupled with higher performance offsets the higher purchase price over the life of the product.
Solid-state disk technology is the clear winner over mechanical disk technology when considering speed, reliability and power consumption. Flash memory does wear out, as it can only support a limited number of writes. Longevity is not an issue for most consumer and enterprise applications, because write leveling technology ensures that writes occur across all of the storage media. It is safe to assume that a write rate of 10 percent in a 24-hour period will not affect the cost of ownership for both consumer and enterprise customers over a period of many years.
SSDs are fast enough that they press the limits of the drive interface. Current SATA disk drives support 3 Gbps interfaces. It is worth looking for SSD disks that support the newer 6 Gbps interface, as the SSD will be able to take advantage of the greater speed.
For those who wish the ultimate in storage performance, look to PCIe interfaces that can perform over a million random disk accesses per second and move billions of bytes of data per second. In fact, many solid-state devices (both SATA and PCIe interface) are fast enough that file lookup by the CPU in the file system can be the rate-limiting step. Only expect to approach a million IO operations per second (IOPS) when performing random operations within a file. Even consumer SSDs will be limited by the CPU during file system operations, such as opens and closes. In other words, SSD storage will be noticeably faster but not instantaneous for many operations, such as booting the computer. Also, some PCIe SSD devices are not bootable, which may pose a problem for some applications.
This leaves price — dollars per gigabyte — as the only area where mechanical disk drives are competitive. Low-end SSD drives cost around $1.50 per gigabyte. Gartner predicts SSD prices will drop to a dollar per gigabyte in 2012. In comparison, mechanical disk drives cost around $0.13 per gigabyte. Switching a 320 GB hard disk drive to a smaller capacity 256 GB SSD disk when purchasing an Apple MacBook Pro will add $650 to the price of a system. This makes purchasing an SSD a luxury for many laptop owners.
Tiered storage systems are where SSD really pays off for enterprise customers. The cachet in caching is that enterprise customers see the performance benefits of SSDs without paying for an all solid-state storage solution. Storage arrays can use solid-state storage dynamically like a large power-safe cache. Active data that creates a “hot-spot” of frequent accesses in the storage array can be transparently moved from mechanical disk to solid-state disk. The requirement is that the active data (e.g. hot spot) fit inside the available high-speed SSD storage. Examples include databases and virtual desktop infrastructures. As the workload changes, these hot-spots can be dynamically migrated on and off the expensive, fast storage to best maintain high storage performance. No human intervention is required!
Tiered storage is also an excellent pathway for growth. As SSD costs drop, a tiered storage system can be upgraded to larger capacity SSD devices. In addition, a tiered storage system can be upgraded to support new, more demanding customers rather than incurring the cost of an entirely new storage array. Similarly, a tiered storage solution will migrate data to slower storage as workloads change and previously “hot” data needs only to be accessed for archival reasons.
A tip when purchasing a tiered storage system: ensure that hot data will move to SSD storage efficiently and quickly so the SSD can accelerate dynamic workloads and those that exhibit “spikey” behavior. In this way, a tiered storage system works to maintain the highest ROI possible and let the enterprise customer cash in on caching!
Happy “cache” computing!
References
1. www.dell.com/downloads/global/products/pvaul/en/ssd_vs_hdd_price_and_performance_study.pdf
2. www.maximumpc.com/article/news/gartner_predicts_ssds_will_cost_1_gigabyte_2012
3. http://news.cnet.com/8301-13924_3-20041259-64.html
Rob Farber is a visiting HPC expert at Irish Center for High-End Computing (ICHEC), supported by Science Foundation Ireland. He may be reached at editor@ScientificComputing.com.
