Four considerations for SSD deployment
- — 02 November, 2009 19:01
While solid state drives offer increased performance, the key to figuring out the role they can play in the data center is balancing that performance against cost.
SSD technology has been around for decades and has become more popular in laptops and desktops, especially among power users clamoring for the biggest performance bang for the buck. The word has spread and businesses are recognizing the advantages of using SSD in the data center to turbo-charge mission-critical applications. Companies can typically use fewer SSDs compared to the number of traditional spinning disks they would have to deploy to achieve equivalent performance.
However, not all solid state technology is built the same. The speed of SSDs doesn't come cheap; the average price per gigabyte is approximately 10 times more than traditional disks. Before adopting SSDs, you need to understand what is driving your performance needs and which SSD configuration is most appropriate for your applications.
Here are four hardware and software considerations for SSDs:
1. The need for speed: Are SSDs the answer?
In its study, "Worldwide Solid State Drive 2007-2012 Forecast and Analysis: Entering the No-Spin Zone," IDC forecasts the market for SSDs will grow 70% between 2007 and 2012, but adoption of this new storage infrastructure will not happen overnight.
SSD provides the power to significantly increase IOPS for the most demanding applications. It also frees up overtaxed traditional drives in tiered storage environments to function at maximum ability. Tiered storage moves data between high-performance, low-capacity drives and slower, higher-capacity drives. However, integrating SSDs into an environment not optimized for the technology can be a lot like fitting a square peg into a round hole.
To balance speed and cost, you should start small without having to purchase a lot of SSDs up front. Make sure you can maximize storage utilization of SSDs by combining the drives with key virtualization applications. See if you can integrate the technology in your existing array or if you need to install another "brick" or array to use SSD. Without the right infrastructure, the drives can quickly become too costly, difficult to manage and inefficient compared with traditional spinning media. When you're evaluating SSDs check with your vendor candidates to determine how SSDs can be integrated into their existing system and how to maximize functionality.
2. Drive vs. cache: Two ways to configure SSD
Once you decide to implement SSDs you must consider the actual configuration. The primary (and much debated) choices for maximizing efficiency are whether you implement the drives as cache or persistent storage.
Cache-based configurations can improve the performance of a server or controller by acting like additional DRAM memory. Many controllers are essentially closed systems and won't accept cache cards, so the only option is to add a dedicated SSD cache, but that approach tends to be costly. Controllers that accept cache cards do so at the expense of a valuable PCI express slot, which would otherwise be used to maximize the number of ports. However, the biggest drawback to implementing SSDs as cache is that it prevents users from tiering storage, which many in the industry believe is the most efficient path to SSD utilization and performance.
Unlike cache configurations, persistent, drive-based implementations let users tier SSDs along with traditional spinning disk. In an automated tiered storage environment, SSDs can be reserved for applications requiring the best performance, which means fewer SSDs are needed. Less-essential data that needs to be accessed quickly can be stored on relatively lower-cost, higher-capacity Fibre Channel or SATA drives, as opposed to being archived off-site or on tape. Automatically moving data back and forth between the solid state drives and disk tiers based on policies, such as frequency of access, can significantly reduce the cost of storing and managing the data.
The decision to implement the drives as cache or persistent storage maps goes directly back to the fundamental question: what do you want to achieve? Cache provides a significant performance improvement for the whole storage infrastructure, and does not require additional software or training. However, caching precludes easy tiering. When SSD is integrated as the top tier in a persistent storage environment, users can purchase only the number of solid state drives required to house the active blocks for their applications. They don't need to purchase SSDs for entire volumes.
3. SLC vs. MLC: It's what's inside the drive that counts
There are two basic flavors of SSD drives: those based on flash memory and those based on DRAM. Since the inception of SSDs, flash has changed the landscape and outpaced DRAM as the chosen data center technology. Although DRAM has performance benefits, flash is significantly faster than disk-based arrays, more affordable than DRAM to implement and is the widely offered format of most storage array vendors.
The flash found in SSDs is further broken into two categories: Single Level Cell (SLC) and Multi Level Cell (MLC). SLC flash is found predominantly in enterprise-class drives and, as the name implies, each data bit is stored in one cell. This format is associated with better reliability, improved longevity and better read/write cycles. MLC drives, while less expensive to manufacture, have slower transfer speeds, higher power consumption and lower cell endurance, and are typically found in consumer memory cards. The bottom line is more data is stored in each cell -- if a cell is lost, more data is lost along with it.
4. Software applications to maximize SSD efficiency
Increasing speed of operation and access to critical applications is the impetus for investing in solid state technology. So once the hardware decisions are made you have to address the software questions. Two storage virtualization technologies noted for their ability to make the SSD performance spike are thin provisioning and automated tiered storage. On top of these, you should also employ storage resource management (SRM) software to automatically track and report how much capacity is being used across tiers, and by the SSDs themselves. The SRM software should provide granular enough detail about utilization to take the guesswork out of SSD capacity planning.
This leads to a question everyone should ask: Are there hidden costs? Many solutions require investment in entire "bricks" and enclosures, which significantly increases the investment. Others allow you to purchase SSDs in smaller increments as the data set grows.
Additionally, will you be forced to predetermine volumes and applications for SSD technology? Can you use thin provisioning with the SSDs, or are you wasting capacity just to allocate storage? Thin provisioning means space is only consumed on these expensive drives when data is written, leaving as much space free as possible and the drives operating at peak performance. While thin provisioning is gaining popularity, few vendors offer the technology for SSDs.
Another consideration is the ability to automate storage tiering. Incorporating SSDs can improve performance, but without the ability to dynamically move data to lower storage tiers, unused data remains static on the high-performance drives. This quickly negates the anticipated benefits. Because industry research shows that 70% to 80% of all data is inactive at any given time, automated tiering helps keep business data online, while removing the need for administrative intervention or data classification software, saving time and money.
By combining thin provisioning with automated tiered storage, users can maximize utilization of SSDs for business-critical applications.
Getting value from SSD
SSD efficacy is based on the ability to increase I/O and utilization, so while the cost per GB may be higher than HDDs, the cost per I/O is far less with flash storage. This reduction is exemplified when looking at usage examples. Some SSD suppliers suggest one SSD can deliver the performance of 30 Fibre Channel drives.
While every data center may not need to process tens of thousands of IOPS daily, implementing SSD technology helps future-proof your environment and position your company to seamlessly handle growth.
Furthermore, the value of SSDs can be measured at the desktop where the benefits of performance and access are experienced first-hand by users. Providing lower latency can improve the experience of running database queries and processing sales reports, meaning the value of SSD technology can be measured both operationally and in business terms.
Fine is director of product marketing at Compellent.