Lower Your TCO With last Solid-State Disks.

Solid-State Disks (SSDs) have existed for more than 20 years, originally reserved for such "high-end" markets as supercomputers and military real-time data acquisition. Historically, the high cost of SSDs has prevented the widespread acceptance of this storage medium in other more cost-sensitive markets. While every data center manager dreams of adding solid-state disks to his data center, financial realities have made RAID devices the overwhelming choice for cost-effective data storage. This is easily understood, since the cost of disk drives has been dropping by 50% annually. Additionally, disk performance has not traditionally been a critical factor in overall system performance. With processor speeds now regularly exceeding 1 GHz, however, it is apparent that CPU utilization is being impacted while high-performance processors and data-intensive applications wait for disks to deliver their data. For many of these demanding applications, adding an SSD would actually lower the total cost of ownership (TCO) when viewed in terms of transactions per dollar.

Standard Storage Architectures

Today's data centers employ two distinct storage architectures. For smaller data sizes and moderate performance requirements, a departmental server provides universal access to a data storage reservoir. For more demanding requirements, a storage area network (SAN) provides an expandable network of storage elements. The main architectural difference is that a SAN's storage resources reside on its own dedicated network (generally Fibre Channel). The departmental server is the more widely accepted of these architectures and is generally simpler and more cost effective than a SAN. A departmental server, however, is quickly overwhelmed by truly demanding storage requirements. The SAN architecture, by comparison, delivers better performance with the potential for virtually unlimited growth in capacity. While both of these architectures could make use of solid-state storage technology, they both depend almost exclusively on mechanical disk drives in one form or another. This dependence has been acceptable because di sks have delivered adequate performance to relatively slow computers and general purpose networks, neither of which could make use of any storage performance increase. Computers and networks, however, have seen dramatic performance increases in recent years. Disk drive performance, specifically access latency, has not experienced a corresponding performance increase. Access latency continues to be the Achilles' heel of disk-based storage, a limitation that is fundamental to the mechanical nature of disk drives. Clearly, today's gigahertz CPUs and Gigabit Ethernet networks require a faster storage technology. The SSD is a perfect fit for this new high-speed world.

SID Background

Solid-state disks have been around a long time and have appeared in many different forms under many different names. The earliest systems were known as mass memory systems (MMS), and were manufactured by a number of prominent companies -- Dataram, Intel, Mostek, Motorola, National Semiconductor, and Texas Memory Systems, to name a few. Most of these companies believed that solid-state storage was the wave of the future and made corresponding plans for tremendous market growth. Each of these companies underestimated the price reduction that disk drives would achieve. The SSD market turned out to be non-existent for all but some I/O intensive military applications. Only Texas Memory Systems (TMS) has remained in this market providing SSDs to these military applications. Recently, the promise of higher volume and widespread commercial acceptance has again returned to the solid-state storage market. TMS, in addition to several other newer companies -- Solid Data, Imperial Technology, Bitmicro Networks -- have be gun to manufacture SSD products to serve this market. This trend truly positions the solid-state disk as the next key component in raising the performance bar in storage networks.

Typical SSD Configuration

Regardless of the manner in which an SSD is used, it is essential that the SSD appear identical to a disk or a RAID from a software standpoint. To meet this requirement, two dominant SSD configurations have arisen: a small form factor configuration designed to replace a disk drive directly and a larger standalone configuration that behaves like a RAID. Some disk replacement configurations actually possess a mounting bracket and a SCSI interface connection which allow them to be installed directly into a computer in a disk slot. Most standalone configurations employ either SCSI or Fibre Channel connections. Both of these methods provide a fast data path between the SSD and the computer. While SCSI connectivity typically only allows attachment to a single computer over shorter distances, Fibre Channel supports multiple computers via Fibre Channel switches and much longer distances. A few standalone SSD systems are available with multiple Fibre Channel links, allowing sustained I/O bandwidth of up to 3,000MB/se c. These SSD units can provide storage capacities up to 64GB, plus options for integrated disk-based backup support. As the SSD continues to experience greater acceptance in the data center, even better SSD products are on the horizon with more features that will have an even greater impact on system throughput.

Emerging SSD Features

The primary motivation behind installing an SSD is its ability to access data in microseconds instead of milliseconds (as with rotational disks). Additionally, SSD bandwidth capacity is limited only by the system designer and is achieved without striping or other techniques that mandate the addition of storage units. By themselves, these features have combined to produce a doubling in SSD sales each of the past few years. With this widespread acceptance, SSD manufacturers are expanding their performance margin with additional features that are inconceivable with standard disk drives. Since the SSD does not incur a performance penalty during random accesses, many users can have simultaneous access to the SSD. Some advanced SSD products support several Fibre Channel ports that can supply an almost unlimited number of accesses without any performance degradation. A standard disk drive could not even hope to attempt this volume of accesses. This high availability SSD could support many computers running a vast n umber of applications. For example: A hurricane is in the Gulf of Mexico and everyone in Houston (TMS's office) wants to know where it is. An Internet weather site would put the hurricane web pages on a SSD with one or more servers accessing this data for the web. Now these web servers will have immediate response to the hurricane data on the SSD allowing twenty thousand Internet users to get their hurricane radar maps every second.

The SSD Cost Advantage

The performance increase of SSDs over RAIDs can sometimes offset their higher price. In these cases, the SSD provides a much faster response time at a lower cost per access. A high performance RAID can support approximately 5,000 random I/O transactions per second (IOPS). Hence a server accessing this RAID will receive 5,000 responses per second to its requests for data. If a data center needs a higher level of service, it must purchase multiple RAIDs. If a storage device could support a greater number of IOPS than that required by the server, then it could be used to support multiple servers. Today's SSDs can handle between 10,000 and 500,000 IOPS and support multiple servers.

A RAID delivering 1,000 IOPS can cost approximately $30,000, depending upon storage capacity. Such a RAID would have a cost of $30 per IOPS ($30,000/5,000=$6). In comparison, a Texas Memory Systems SSD (SAM-520) with a Fibre Channel link delivers 50,000 IOPS at an approximate cost of $50,000, depending upon storage capacity. This places the SAM-520's cost at only $1 per IOPS. While both systems cost approximately the same ($30,000 versus $50,000), the higher performance of the SSD makes its TCO only a fraction of the TCO for the RAID. This analysis is based exclusively on dollars per performance and not the dollars per gigabyte metric preferred by manufacturers of rotational disks. Also, SSDs do not have many mechanical components and are therefore less subject to mechanical wear. SSDs can decrease your TCO, lower maintenance costs, and be managed easily.

Typical SSD Applications

Traditionally, only the most demanding applications required or justified the added cost of an SSD. As the performance requirements for storage continue to increase, however, the SSD is finding its way into many mid-level computer applications. Additionally, the widespread acceptance of the SAN architecture provides an excellent opportunity for fast SSDs to have a dramatic impact on overall system performance. An SSD adds an immediate speed benefit and does so for a significantly lower cost per I/O than a RAID. It can be attached to a single computer via a SCSI connection or to many computers via Fibre Channel connections. Many system administrators find that simply connecting an SSD to a SCSI port reduces server I/O wait time significantly.

As more connections are made to the SSD, more available bandwidth is utilized and additional I/O accesses are completed within the same time frame. Several obvious SSD applications include web servers, database servers, shared file storage, index file storage for file system management, and other hot file retrieval operations. Hot file and index file storage are ideal SSD applications because they utilize the strengths of an SSD in conjunction with those of a RAID. Frequently accessed files may be stored on an SSD, when performance is critical. Larger and less frequently accessed files may be stored on a RAID, where file sizes warrant greater storage capacity. An additional benefit of this approach is that RAID accesses are decreased as SSD accesses are increased, thereby reducing mechanical wear on the RAID.

SSD storage has finally found its way into mainstream computer and network architectures. While it is not ideally suited for every application, it is a perfect match to the requirements of high-performance applications. Of particular importance is an SSD's ability to make other existing storage components more efficient. RAIDs benefit greatly from the addition of even minimal amounts of solid-state network storage. While the overall use of SSDs may be relatively small today, rapid increases in network and CPU performance will only increase the role of SSDs in the future -- leading to more efficient storage systems with higher performance and a lower total cost of ownership.

Holly Frost is the CEO and founder and Charles Camp is VP of Engineering at Texas Memory Systems (Houston, TX).