Solid-State Drives Redefine Storage

Original Article Date: 2009-01-15

Intel enters the hard drive market, and not a spindle in sight! 

For permanent storage, computers have been using the same type of device since IBM invented the magnetic hard drive in 1956. And in a more modern setting, hard drives as we know them today (in compact form factors that fit into micro-computers) have been in use since Shugart ("Seagate") Associates introduced the SA1000 5MB 8-inch drive back in 1979.

So we've been using the same technology of a magnetic disk platter to store all our data in our PCs for nearly 30 years. Whilst capacities have increased 300,000 times in this time (hard to believe, that), and form-factor sizes have reduced, the basic design of the hard-drive has remained largely unchanged. And with this basic design has come a number of limitations, which center around the fact that, at its heart, the hard-drive is an electromechanical device.

"No moving parts" is a good motto for component reliability. Unfortunately, the hard-drive has at least two - the spindled platter, and the read-head armature. Disk failures are a fact of life, and whole systems are built around this fact (RAID architectures, elaborate backup software etc.). And the delicate design of the spindle and read-head does not like to suffer kinetic shocks. Forget about dropping a hard drive, or the computer it's contained within.

Additionally, the mechanical aspect provides a certain ceiling of performance, compared with the seemingly limitless domain of solid-state electronics.

With these limitations, and a clear acceptance that the design of the hard drive is imperfect, it's somewhat surprising that an alternative to persistent computer storage has taken so long to be realized.

But that alternative has now arrived, in the form of solid-state drives. You've probably heard of them - in fact we've likely all used them in the form of memory for our digital camera or a USB thumb drive. But it's taken until this time for such technology to really mature to the point that it is a viable alternative to electromechanical hard-drives, particularly in the server and workstation market.

Storage on Flash Memory

There are in fact two types of solid-state drives (or "SSDs"):

• Those that use DRAMs, like conventional memory. These are fast, but extremely expensive, and require some kind of power to be maintained at all times, otherwise the data is lost.
• Those that use Flash memory, which is non-volatile, meaning that the data is persistent when all power to the memory is turned off. These are the most popular devices that have evolved, and hold the most widespread commercial promise. This article is about this type of SSD.

SSDs are not new. They have been around since before the 1980s, and have found niche applications such as the military or aviation where extreme ruggedness and very low failure rates are required. But it's only recently that SSDs have evolved in terms of performance, capacity and affordability to make them viable to the wider enterprise and consumer market.

Laptops were the first to begin to make use of commercial SSDs, since the high kinetic shock resistance of a drive with no moving parts made them particularly suited to environments that can be subject to heavy accelerations through being dropped or simply moved around frequently.

It's only been in the last few months, however, that the enterprise space has found a use for such drives. This is primarily due to the same or even inferior sustained read/write performance of SSDs when compared to conventional hard drives, despite the much higher cost in $/GB. As servers and workstations aren't moved around too much, there was no interest in a drive that was more robust against kinetic shock.

There was however, interest in the higher reliability of these drives, expressed in MTBF (Mean Time Between Failure Rate), as well as the much lower latency (or seek time) available due to the lack of any moving parts. These last two benefits led a select number of manufacturers to push development forward on this technology so that the twin benefits of higher performance and reliability could be offered.

Flash memory is available in two basic variants - Multi-Level Cell (MLC) and Single-Level Cell (SLC). Without getting into the technicalities of what these mean, you should simply know that MLC is cheaper but not as high-performing or as reliable as SLC. So when looking at an SSD, it's important to know whether it is built using MLC or SLC.

One final consideration that has concerned some consumers is that the NAND chips within Flash can only be written to so many times before they fail. If SSD memory was written in the same way as a hard-drive, this would mean some chips would be written to much more frequently than others, causing them to fail perhaps within the expected lifetime of the device. This problem has largely beeen overcome by wear-levelling, in which the controller constantly moves around frequently written-to sectors, so that all chips are used, and the lifetime of the device is (so manufacturer's tell us) extended beyond anyone's concern in even the most data-intensive environments.

Pros and Cons of SSDs

Advantages

• High kinetic shock resistance
• Very low latencies
• High sustainable read/write bandwidth
• High reliability due to no moving parts
• Very low power draw when inactive
• Near-silent

Disadvantages

• Cost per gigabyte is an order of magnitude above that of conventional HDDs
• Limited capacities

The above points highlight the overall advantage SSDs enjoy over conventional drives, with perhaps the greatest single advantage in the server and workstation market being the low-latencies. It's important not to get too focused on headline sustainable read/write times when comparing with conventional HDDs. Remember that, owing to the lack of moving parts in SSDs, latencies are extremely low, being at least an order of magnitude lower than conventional hard drives. This factor becomes crucial when dealing with a large number of small block random reads and writes, which is typical in a server application, and especially relevant in databasing.

So, often, a better guide to performance in server computing is to consider IOPS (Input/Output operations Per Second), as long as the benchmark states clearly what block size is being used. This is because larger block sizes will deliver lower IOPS proportionately and smaller block sizes will yield higher IOPS scores. 4K random read and 4K random write are commonly used in IOPS benchmarks. SSDs shine particularly well with IOPS, with performance being 20-100 times better than conventional hard drives.

SSDs draw much less power than conventional hard drives, usually only a fraction of a watt when idle, compared with 8-10 watts for their electromechnical equivalents. They are also near-silent, which is attractive to desktop and workstation users tired of hearing the background whine of a high-RPM drive spindle.

Not all SSDs are the Same

Fusion-IO blurs the line between RAM and HDDs with the ultra-fast ioDrive 

Products now currently available from two vendors, Intel and Fusion-IO have caught my interest, since they are specifically targetted at the server and workstation markets.

Let's first consider the Intel SSDs. They come in two familes, one with the superior SLC and one with MLC. I actually had the MLC version integrated into a system recently, and it performed very well indeed. We proved out Intel's claim of sustainable read times being 250MB/s, which is basically the upper limit on the SATA-II 300 bus! Windows installed and booted in a fraction of the time normally taken. And the tiny drive was completely silent!

I am recommending the Intel SSDs over others because they are aimed at the enterprise market, and whilst they're more expensive than the competition, this is justified in higher performance. Intel have developed these devices from the bottom-up to be optimized with the SATA-II bus, whereas competitors are still using older parallel IDE interfaces, which are then coverted to Serial ATA, resulting in performance bottlenecks.

Fusion-IO have altogether abandoned the SATA-II interface, and gone straight for a "drive" that plugs directly into the PCI-Express bus via a PCIE*4 connector, like any other add-on card. The advantage of doing this is there is no 300MB/s restriction on speed imposed by SATA, and it allows this "drive" to perform incredibly, with sustained read/write speeds of 700MB/s and 550MB/s! Such speeds have only been previously possible using RAID with 8+ drives! These claims have been backed up by independent benchmarkers. There is a hefty price tag for these Fusion-IO drives, however, with the 80GB version at around $2,900, and the 320GB version at around $14,000.

The following table summarizes the performance, reliability and economic characteristics of these SSDs, comparing them to SATA and SAS conventional HDDs.

These drives are available now, and can be integrated into the workstation or tower server of your choice. The only limitation is, at this time, the form-factor of SSDs prevents them being mounted in hotswap assemblies. But then, with their far-superior reliability, why would you need to hotswap?

Summary

In comparing SSDs to conventional HDDs in the above discussion, it's clear that they have the overall advantage. The only real downside is currently cost, but with anything still relatively new, this is always an issue, and it is pretty much accepted that as the new technology moves into the mass-market, prices become but a fraction of what they once were.

With the expected lowering in cost of SSDs as they become more prevalent, I expect that they will replace conventional drives in most applications over the next 5-10 years. The only exception might possibly be in specialised applications such as video surveillance or media serving, where very large capacities are still required. But even then, at some point in the future, with more R&D moving over to SSDs from conventional drives, I expect SSDs to overtake conventional drives on capacity.

And then... one day in the future there will be a tech article somewhere with a photograph of the last drive spindle to be milled before it is mounted onto a magnetic drive platter. Good-bye electromechanical, and hello solid state!

Best regards,

Ben Ranson
Chief Systems Engineer
Electronics Nexus
http://elnexus.com
ben@elnexus.com
1-877-773-5366