SATA Overtaking Storage Arena in Record Time
Sometime in the third quarter of 2006, Serial ATA became the dominant storage interface, accounting for more than half of all drives sold in that time period, and for an overwhelming majority of total disk storage in terms of raw gigabytes. Moreover, SATA has become the primary storage interface for most new systems, with Parallel ATA, or IDE, drives and interfaces being relegated to "legacy" and non-PC uses (such as media systems and gaming consoles), while SCSI's dominance over enterprise servers and storage systems is also being challenged like never before.
On the surface, this may appear to be nothing more than the natural progression of a planned industry evolution, but it's a fairly impressive accomplishment given the relative newness of SATA technology. According to Knut Grimsrud, president of the Serial ATA International Organization responsible for managing the SATA standard, the 1.0 specification was published in August of 2001, while the 1.0a update came out in January of 2004. That means we've gone from first specification to clarification to primary interface within a mere five years.
By comparison, USB has been around for about 12 years now, but most PCs are still sold with legacy connectors for PS/2 devices, serial and parallel devices, joysticks, floppy drives and more, almost all of which were expected to have been made obsolete by now, and there are still large chunks of the add-on market that continue to give preference to the legacy interfaces. From that perspective, SATA's rapid and broad adoption rate is almost spectacular. It may even be unprecedented—whereas technologies like Accelerated Graphics Port (AGP) and PCMCIA were also adopted very quickly, they were typically constrained to specific platforms, and did not see universal adoption across all sectors. (There were plenty of servers that had neither, for example.) Conversely, SATA is meeting with rapid success just about everywhere.
Fun With Numbers
The most obvious example of this can be found in the desktop PC arena, where SATA is a natural replacement for older PATA drives and interfaces. A couple of years ago, most of the motherboards in this class only had a couple of SATA connectors, with just a few of them having four or more. With the introduction of new CPUs from Intel and AMD, however, the industry has seen a wholesale refresh of the product lines, and almost all of the new PC motherboards have a minimum of four SATA connectors (some having as many as eight or more), and many of these boards now only have one PATA interface (and some don't have any PATA connectors at all).
All told, 66.7 percent of the hard drives that were sold in Q3 2006 in this space were SATA, according to John Rydning, research manager for hard disk drives and components at IDC. While legacy PATA drives still account for a huge chunk of sales (constituting the bulk of the remaining 33 percent), the simple fact is that SATA is already outselling PATA by a 2-to-1 margin in the desktop sector, and that's within just a few short years of general availability.
Industry insiders said they expect laptops to switch to SATA even faster than desktops, since replacement cycles within that sector are usually more compressed, and because the smaller number of laptop vendors and models make every change more noticeable. Sure enough, many of the new laptop models released for the new Intel and AMD processors are also now using SATA, and according to Rydning, 44 percent of the 2.5" laptop drives sold in Q3 were SATA. That figure may seem small compared to the desktop share, but consider that SATA only accounted for 18 percent of the laptop drives sold in Q1, and you can see how quickly the transition is taking place.
SATA is also encroaching into multi-user storage, with many of the entry-level and mid-tier servers and external storage systems taking advantage of its relatively low cost-per-gigabyte entrance fee and relatively high performance characteristics. For example, many of the server-class systems that have been released over the last couple of years have adopted SATA as the primary interface for local storage (just as with the desktop sector), but those interfaces are also frequently used for less-demanding applications such as serving up static content. Meanwhile, scores of new vendors have entered the storage area network (SAN) and network attached storage (NAS) markets over the last couple of years, and many (if not most) of them are selling cabinets based on SATA storage designs.
By most measures, SCSI still offers better performance than SATA (especially in multi-user applications), and Serial Attached SCSI (SAS) is positioned to maintain the performance advantage for the next few years as it takes over from parallel SCSI in new systems. However, the same technology that makes it faster also saddles SAS devices with a cost premium that is several multiples of the cost-per-gigabyte of SATA, and this presents a real challenge to SCSI. Cumulatively, SATA is expected to continue doing well in the bottom parts of the multi-user markets, but it's expected that SAS will eventually come to dominate the rest of that space.
John Monroe, vice president of research for storage markets at Gartner, says that SATA is currently outselling SAS in multi-user markets by a 2-to-1 margin, but he also says that traditional parallel SCSI and Fibre Channel are each beating SATA by that same 2-to-1 spread. His projections show that SAS volume will grow to fill the traditional high-end enterprise SCSI market within the next two years, and that while SATA will continue growing, the bulk of that sector will be firmly behind SAS. He also believes that SATA and PATA already account for more raw gigabytes of storage than SCSI-based systems (due to the greater storage capacity of ATA drives over SCSI), but that this will not seriously impact SAS installations in the high-end enterprise sector, since the bulk of the unit volume is being driven by a handful of vendors.
IDC's Rydning suggests essentially the same thing, as does Jonathan Gerber, vice president of trading for Horizon Technology, a distributor that handles hard drives from multiple vendors. According to data published by Horizon, SATA accounts for approximately 70 percent of all current unit volume, and that same report also cites softness in the premium-priced SCSI drive market. More recently, Gerber maintained that high-end enterprise drives would likely stay bound to SCSI and SAS due to performance requirements.
My feeling is that SATA will take a larger chunk of this sector than expected. While SCSI is demonstrably faster and more powerful than SATA, many systems have used SCSI simply because it has historically been the only real option available, and not because of its performance characteristics. (The number of cheap and slow SCSI controllers out there provide much support for this argument.) But with the availability of moderately fast SATA drives with multi-year warrantees and features like native hot-swapping and command queuing, SCSI is no longer the only game in town for server-class storage.
Furthermore, given the number of vendors that are investing heavily into pure-SATA market plays, it seems likely that performance and reliability will only improve, meaning that it will become more suitable for more people as time progresses. I certainly don't expect SATA to completely eliminate SCSI anytime soon, but it seems likely that the low cost per gigabyte in conjunction with a dynamic vendor community is more apt to favor SATA in the long run.
Performance And Features
The reason for SATA's rapid adoption is pretty simple: SATA delivers a lot of valuable features over the older PATA technology, but doesn't cost much more. In fact, there's no price difference at all in the desktop sector.
The most obvious technical advantage is that the higher-speed serial interface allows for much faster transfers. Serial interfaces only require a simple transmit and receive loop, which can be run as fast as the system can process the data. Conversely, parallel interfaces require clock-synchronization across multiple pins, and it becomes increasingly difficult to maintain even clocking as clock speeds get higher. Parallel ATA drives have managed to become incrementally faster over the years, topping out at a theoretical limit of 133 MB/s, but SATA already supports speeds up to 300 MB/s, and faster rates are coming.
Although SATA drives that can consistently fill that kind of bandwidth are not yet available, the higher ceiling does allow for faster transfers when the data is already in the drive's cache, which means that SATA has the ability to "burst" data a lot faster than an identical PATA drive.
Another benefit from this simple change from parallel to serial interfaces is that the higher clock speed also provides snappier command and response exchanges, so more complex operations are possible within the same time interval. The simpler design also allows for smarter devices that can spend more time managing the disk and commands, and less time managing the interface, which allows for smarter drives.
And since the speed is just an interface cap, you can also lash multiple independent serial channels together and get very good levels of cumulative throughput by running multiple drives at maximum load. This principle was highlighted in my review of Enhance Technologies' Q14 2.5" drive array, which delivered pretty surprising results when all of the drives in the RAID array were pushing data together. Similarly, a simple RAID array of desktop-class SATA drives can easily stream large data transfers faster than a gigabit Ethernet network can handle. This is one of the reasons why SATA is so popular with the new crop of NAS vendors—cheap drives are no longer the limiting bandwidth factor.
SATA also provides native support for important features like hot-swapping and command queuing (although they still have to be implemented in the devices, which is not guaranteed with lower-cost units). These features allow the technology to be considered practical for higher-end applications that would otherwise be out of reach for PATA. The thinner and longer SATA cables also allow for better airflow management, which helps to keep today's faster computers running cool.
One benefit that isn't readily apparent, but which has proven to be useful here, is that SATA devices have a common physical connector, which makes resource management somewhat simpler. For example, I can pull a spare from a server array and swap that into a dead desktop system (or vice versa) with no difficulty whatsoever, while the same simply wasn't possible when my desktops were using PATA and my servers were using SCSI. This principle also applies to 2.5" drives (albeit to a lesser extent), in that the common interface means that I don't need special controllers for those drives, so I can do things like clone a laptop drive on my desktop PC with no special hardware or cabling. (There can be some differences in the way that laptop drives handle their LED, but that's a minor penalty.) Along the same lines, SATA also shares a common physical interface with SAS, and SATA drives can (theoretically) be used on SAS controllers. Thus, an easy way to knock $500 off the price of a blade server is to use a SATA laptop drive instead of the more expensive (and perhaps unneeded) SAS drive. The use of a common interface has already simplified some of my recovery operations, and it's already one of my favorite features, even though it isn't one of the common selling points.
Another of SATA's unstated features is the renewed competition that it has spawned over the last few years. SATA is a new technology, so innovators have been presented with a chance to compete on the same footing as older vendors who were established on other architectures, but who do not yet have brand dominance with SATA buyers. Furthermore, the numerous performance and feature benefits that are inherent in SATA's design pretty much apply to everybody, and the only real difference between the players is in their strategy and implementation.
For example, vendors like 3Ware and Areca are pushing SATA performance as far as they can to try and corner the top of the market, while vendors like HighPoint are trying to compete at the low-cost sector with controllers that use device drivers to off-load RAID calculations to the host CPU (thereby eliminating the cost of an onboard dedicated processor). Some vendors, such as Intel, are attacking multiple sectors all at once, with low-end hardware-assist RAID chips, like the ICH7R being sold to motherboard vendors, and feature-rich add-on cards being sold to end users simultaneously. Meanwhile, disk drive vendors like Western Digital are trying to develop the fastest drives possible, while vendors like Seagate are trying to flood the zone with drives for every lifestyle. This rush of innovation and capital investment has resulted in some amazing technology and deals.
On the other hand, the newness of the sector has also resulted in some gaping holes. For example, storage devices other than disk drives have been slow in coming, and can still be hard to find—there are only a handful of DVD burners that have SATA interfaces, and the number of tape drives with SATA interfaces is also quite limited.
Another problem that has followed from the immaturity in this space is that some implementations are still missing essential functionality that we take for granted elsewhere. For example, the number of controllers with drivers in the Linux kernel can be counted on one hand, and the rest of them either require special builds or do not have workable drivers available at all. For many users, this means that running Linux on a SATA RAID can only be classified as experimental at best.
Just as a simple data point here, the last three server motherboards I've looked at have all included Intel's ICH7R SATA RAID controller on board. Intel provides Windows XP drivers for the controller on its Web site, and it used to provide drivers for Linux 2.4 kernels, but has since made the strategic decision to defer to dmraid for Linux 2.6 kernels. However, dmraid is still only at 1.0 release-candidate status, and currently only supports RAID-0 and RAID-1 on the ICH7R. Due to limitations with the GRUB boot loader, it's not currently possible to actually boot a Linux 2.6 system off an ICH7R RAID controller either. Bottom line: if you've got an ICH7R and you want to run SATA RAID features under Linux, then you'll have to settle for software RAID, or you'll need to start shopping for another controller, because the free one probably isn't going to work. These problems are common with other controllers that don't provide Linux kernel drivers too, and this phenomenon is by no means limited to Intel.
In spite of these difficulties however, it's clear that the performance and features of SATA are plenty good enough for most applications. For these reasons, I've standardized on the technology for all of my systems, and have replaced almost all of my other storage systems within the last year. It's likely that I will acquire a couple of high-end servers that will use SAS, but those will definitely be the exception, if it happens.