By Dave Simpson
Serial ATA (SATA) disk drives took the market by storm last year, and the popularity of high capacity (400GB per drive), low-cost SATA subsystems is expected to continue unabated this year. For example, International Data Corp. (IDC) expects shipments of SATA disk drives to jump from 1.6 million units last year to almost 3.6 million units this year (see figure)-and that’s not including desktop shipments.
The next advance in SATA, SATA II, is coming later this year, and it will address some limitations of SATA, most notably relatively low performance and, to a degree, reliability.
It’s still unclear when the SATA II subsystems will reach the end-user market, with estimates varying from mid-year for arrays from smaller vendors to the end of the year or 2006 for SATA II subsystems from the larger vendors (which have longer qualification cycles for components such as disk drives, controllers, and enclosures).
(Disk arrays based on Serial Attached SCSI [SAS] drives, which are both competitive with and complementary to SATA, are expected to lag SATA II subsystems by a couple of quarters, according to most estimates. SAS enclosures will be able to accommodate SAS and/or SATA drives, enabling OEMs, integrators, and end users to mix and match drive types according to performance and cost requirements.)
SATA II functionality is expected to come in two phases, with the first phase marked by subsystems with features such as native command queuing (NCQ), hot-swappable drives, port selectors, port multipliers, enclosure management, and enhanced backplane support. It’s important to note, however, that some vendors have already provided some of these features in their first-generation SATA systems, and not all of the features are necessary for a sub-system to be labeled SATA II. In the case of some of the features, SATA II will take the functionality from the subsystem level to the drive level.
SATA II’s biggest performance advances come from NCQ, which will be available with most SATA II arrays in the first phase of delivery, and from a doubling of bandwidth from SATA I’s 1.5Gbps (150MBps transfer rate) to SATA II’s 3Gbps (300MBps).
NCQ algorithms optimize the order in which disk drives execute commands (as opposed to executing them in the order they are received, which leads to mechanical overhead). In other words, NCQ enables a disk drive to take multiple requests for data from the processor and re-arrange the order to maximize throughput. (For more information on NCQ and other SATA II features, visit www.sata-io.org.)
SATA II’s NCQ is similar to, although not as advanced as, the command queuing (or tagged command queuing) available with Fibre Channel and SCSI disk drives. NCQ promises significant performance improvements, but at this time it’s impossible to quantify because the drives are still in testing mode. Also, more importantly, performance is very much dependent on the type of I/Os (small blocks or large blocks) required for the application, as well as factors such disk rotation speed, queue depth, nature of the I/Os (e.g., “bursty” or streaming), etc.
However, some vendors are providing ballpark numbers on the performance improvements users might get from NCQ. For example, Tracy Bowden, product manager in LSI Logic’s RAID division, says that on average NCQ may provide a 25% to 40% performance improvement over SATA I at the drive level assuming small I/O (4KB) workloads that are typically associated with transaction-processing database applications. (LSI’s SATA II RAID adapters were expected to go into production this month.)
Guy Carbonneau, VP of engineering at nStor, which is shipping SATA II-ready subsystems, generally agrees with those estimates.
He estimates that, assuming very I/O-intensive database applications with small block sizes (4KB or smaller), users might expect a 30% to 40% performance increase from NCQ.
Performance increases from NCQ will be far greater in applications with small block sizes compared to applications that use large block sizes such as video servers, backup, etc., where the performance increases may be minimal or even zero.
Paul Vogt, director of Adaptec’s components business, says that the performance improvements from NCQ could be anywhere from 10% to 70% and that users may realize as much as a 60% to 70% performance increase in some database applications. (Adaptec is currently shipping SATA II-ready controllers.)
SATA II will also double SATA’s transfer rate to 300MBps, although disk arrays based on drives with this speed aren’t expected until at least late this year or 2006.
SATA II arrays will also address some of the reliability issues associated with SATA, although it’s important to note that reliability has more to do with how the disk drives are manufactured than with the SATA II specification, which only defines the disk drive interface. Basically, reliability comes down to how much cost a disk drive manufacturer is willing to build into a drive.
Among the reliability features to look for in SATA II subsystems (some of which have been available in SATA-I arrays in some form) are the following:
- Support for hot-swappable (or hot-pluggable) disk drives, which means that users don’t have to bring a server down to replace a failed drive;
- Port selector technology, which is similar to the dual-port technology available on Fibre Channel and SAS drives. However, SATA II’s version uses an active-passive fail-over implementation vs. the active-active fail-over implementation (where both controllers can talk to a drive at the same time) in Fibre Channel and SAS disk drives;
- Enclosure management, which is based on the same protocols and standards (SES and SAF-TE) used for enclosure management of Fibre Channel and SCSI systems; and
- Improved backplane support.
Another feature to look for in SATA II subsystems is port multiplier technology, which enables a single controller port to fan out to 15 devices (although four to eight drives will be more practical for performance reasons). For example, an 8-port controller theoretically could support 120 (8x15) disk drives, enabling OEMs and integrators to increase the number of drives in a SATA enclosure while decreasing cabling hassles.
Few would argue that SATA II will put SATA in the same league as Fibre Channel, parallel SCSI, or SAS when it comes to reliability and performance (exception: Western Digital, with its SATA-based Raptor disk drives, argues that SATA is on par with Fibre Channel and SCSI in many applications), but it does tighten the race a bit. On the plus side, users have more disk drive interface choices; on the minus side, making the right choice becomes more difficult.
“SATA only gets dinged on reliability and performance when people use the systems in applications that are better suited for other interfaces such as Fibre Channel or SCSI,” says Stan Skelton, director of strategic planning at Engenio Information Technologies, which supplies disk arrays to OEMs such as IBM, StorageTek, and Sun.
Skelton notes that the old practice of associating disk drive characteristics such as performance and reliability with the drive interface is passé. “Drive manufacturers are separating interface [attributes] from drive [attributes],” says Skelton. “For example, SATA no longer necessarily means lower reliability.”
So far, SATA subsystems have been used primarily in nearline, or secondary, storage applications such as disk-to-disk backup, but the next generation of SATA arrays may wind up in enterprise, or primary, storage applications. “Initially, low cost was the key requirement for SATA, but improvements like SATA II will take SATA into more enterprise applications,” says Sean Chang, president and CEO of start-up Rasilient Systems, which specializes in SATA RAID subsystems.