By Jim Handy
-- Although solid-state disk (SSD) drives have seen insignificant adoption in the PC market, considerable excitement exists in the enterprise space for these devices.
The key reason is speed. Although the price per gigabyte for an SSD drive is prohibitive in comparison to that of a traditional hard disk drive (HDD), there are server applications that use large numbers of HDDs at a fraction of their capacity to increase I/O bandwidth. In many cases, an SSD can provide more speed at an adequate capacity for a reasonable price.
The difference between an enterprise server and a PC is that enterprise servers use a number of HDDs, whereas a PC uses only one. In a multiple-HDD system, mass storage is ranked into a storage hierarchy, with faster, more costly storage devices being placed in small amounts nearer to the processor and slower, cheaper storage devices used in larger capacities further from the processor. In today's systems, the more expensive storage devices consist of enterprise-class HDDs -- disks with a high I/O rate because of high spindle speeds, wider tracks, and faster, more expensive interfaces such as Fibre Channel and SAS. Slower bulk storage devices are typically based on the same low-cost HDDs that are prevalent in PCs. This is the storage hierarchy typical of all computers, which extends all the way from processor caches down to archival storage.
An example of this hierarchy is shown in the figure, which gives a rough idea of where the different elements of the storage hierarchy fit from the perspectives of bandwidth and cost per gigabyte. We use a log-log chart format to allow us to see all the data that would be hidden if we were to use a linear scale for either performance or cost. The three orbs labeled L1, L2, and L3 are three possible layers of cache in or around the processor.
Since NAND's price per gigabyte has fallen below that of DRAM in recent years, system designers have found interesting ways to tap into NAND technology to improve performance while lowering costs. Flash SSDs are one means of reaching that goal.
Enterprise-class HDDs fit at the top end of the HDD oval in the figure. As flash-based SSDs move into this market they pose a significant threat to enterprise HDDs, and many OEMs and IT managers expect future systems to be built using a combination of SSDs for speed and low-price HDDs for mass storage, eliminating the enterprise HDDs that might otherwise be used between low-cost HDDs and DRAM.
Flash SSD manufacturers have been working over the past few years to produce units that satisfy the enterprise server OEMs' needs at a satisfactory price. Their first offerings were expensive (about $10,000) so these devices are currently are being used to replace the most costly types of HDD-based enterprise arrays: those comprising short-stroked drives.
What is short stroking?
Short stroking is an approach to achieving the maximum possible performance from an enterprise HDD. The technology is relatively common, although the proportion of the overall enterprise HDD market that is used in this way is small. Objective Analysis estimates that less than 10% of all enterprise HDDs are short-stroked. The figure on p. 29 illustrates the basics of short stroking.
Two causes of delay in an HDD are access time and rotational latency. Access time is the time required for the disk's head to find a requested track, often referred to as "seek time," and depends on the distance from the current track to the requested track. Rotational latency is the time it takes for the requested data to move under the head after the head has found the right track.
While a programmer can do little to improve the latency, other than to use high-RPM disks, programmers can ensure that the head motion, and thus the access time, is as small as possible. This is done by using only a few adjacent tracks on the disk and completely ignoring the rest of the disk -- a process referred to as "short stroking."
While a short-stroked drive will only access a fraction of the available disk space, the data will be read off the disk at a significantly higher speed than normal. A disk with tens of IOPS can be "coaxed" into providing data at a few hundred IOPS by using this method. In some cases, users find that this is a worthwhile trade-off.
Another trick programmers can use to accelerate I/O is to use only the tracks at the outer edge of the disk. Data on the outer tracks is accessed at a higher rate, so transactions on these tracks are significantly faster.
An example of this approach is IBM's high-end Tier-1 storage system: the DS8300 Turbo. Offering 123,000 IOPS and a maximum latency of 16ms, this system includes 512 HDDs, in a mix of 73GB and 146GB capacities, which are mirrored and configured in a RAID formation.
Because of the redundant data required, and the fact that the drives are short-stroked, the system's 53TB of internal storage capacity whittles down to only 9TB of usable space.
An SSD can often support tens of times the IOPS of a short-stroked HDD. Often the higher bandwidth of the SSD, in tandem with the very small capacity actually used in a short-stroked HDD, will provide an opportunity for an SSD to replace a bank of HDDs. As long as the SSD's capacity is as great as that used in the short-stroked HDDs, and as long as the SSD's bandwidth matches that of the HDD array at a competitive price, the SSD may provide a more cost-effective alternative to an array of HDDs. This is the case in most of today's deployments of SSDs in enterprise environments.
Objective Analysis believes the market for short-stroked enterprise HDDs will be the first market to completely convert from HDDs to SSDs.
SSD market forecasts
Objective Analysis has arrived at estimates for the enterprise SSD market through two unrelated forecasting techniques. One of our forecasts uses the enterprise HDD forecasts of Coughlin Associates and applies some judgment to the numbers. This forecast results in strong similarities to the "bottom-up" forecast we derived by using a thorough analysis of each enterprise application type.
Our bottom-up forecast finds that the enterprise SSDs will first be adopted in transaction processing systems, but over the long term even stronger growth will occur in large Internet systems.
Our top-down forecast is based upon the following assumptions:
--5% to 10% of enterprise HDDs are short-stroked. This estimate is based on conversations with many parties in the industry;
--Acceptance of enterprise SSDs will be slow at first as IT managers evaluate the risks of introducing this new technology;
--SSD prices will decline with the conversion to MLC NAND flash technology and as SSD designers find ways to substantially reduce other manufacturing costs; and
--Today, arrays of 10 or more short-stroked HDDs can be economically replaced by a single SSD. As enterprise SSD prices decline, one SSD may economically replace even fewer enterprise HDDs.
This gives us a unit shipment forecast that grows from almost negligible shipments today to 1.7 million units in 2013 -- an average annual growth rate of nearly 150%.
Our revenue forecast is driven by this unit shipment forecast. It indicates that enterprise SSD revenues will grow at a strong 67% rate through the forecast cycle, based upon very strong 148% unit growth somewhat offset by an average annual price decline of 40%.
Enterprise SSD unit shipments will grow in 2013 to nearly 100× the expected shipments for 2008. In 2013, SSD revenues are expected to exceed $1 billion.
Enterprise HDDs are threatened by this new technology, which will initially replace enterprise HDDs at a 10:1 ratio, dropping to 3:1 by the end of the forecast period. This means that the enterprise HDD market will shrink faster than the enterprise SSD market can grow.
The enterprise server market is warming up to SSDs faster than has happened in the PC market, and with good reason. Today, entire arrays of costly high-speed enterprise HDDs can be economically replaced with a single enterprise-class SSD. Although enterprise SSDs are anything but cheap, at close to $10,000 each, they can often be used to replace an array of 10 or more HDDs, providing more speed in a much smaller footprint for about the same price.
Objective Analysis forecasts that SSDs will rapidly replace a growing portion of the enterprise HDD market as the technology matures, allowing prices to fall.
Jim Handy is the director of the Objective Analysis semiconductor market research firm. This article was excerpted from the recent Objective Analysis Report, Solid State Drives in the Enterprise.