Text preview for : Serial ATA - A Promising New Alternative for Enterprise Storage Applications.pdf part of Western Digital Serial ATA - A Promising New Alternative for Enterprise Storage Applications Western Digital Serial ATA - A Promising New Alternative for Enterprise Storage Applications.pdf
Back to : Serial ATA - A Promising | Home
Serial ATA
A Promising New Alternative for Enterprise Storage Applications
O VE RV I E W / E XE C UT I VE S UMMARY
For desktop and portable computers, the Parallel Advanced Technology Attachment (ATA) interface is currently the most popular
protocol for moving data between the hard drive controller and system memory. Designed to provide a "best of both worlds" solution,
Serial ATA (SATA) replaces the existing parallel bus with serial links, and adds features designed to increase the performance, reliability
and scalability of ATA-based devices while retaining ATA's significant cost advantage.
B AC KGR OUN D
As an evolutionary replacement for Parallel ATA, a consortium of industry leaders developed a new storage interface specification called
Serial ATA 1.0. This new specification was formally introduced in August of 2001. Serial ATA II, a second iteration of the specification
for SATA servers and network attached storage, was published in October 2002. It will allow vendors to further address the needs of
networked storage segments via enclosure management, backplane signaling, and cabling and performance improvements.
Traditionally, ATA has not supported the requirements of enterprise systems. Traditionally, enterprise storage applications have been
dominated by solutions based on the SCSI (Small Computer System Interface) standard, with the more recent entry of Fibre Channel.
Systems based on these technologies deliver the high performance and reliability required for mission-critical applications. However, the
cost for systems based on SCSI or Fibre Channel can be substantially higher than technologies based on ATA.
Storage vendors quickly realized that Serial ATA has distinct advantages over traditional ATA and parallel SCSI that make it a viable
alternative for many enterprise storage applications, enabling storage systems at what current pricing models estimate to be one-third the
cost of SCSI-based systems. The result is a cost-effective alternative to SCSI technology for all but the most critical direct-attach and
networked storage applications.
S E R I AL ATA F E A T U RE S AND B E NE F I T S
While SATA represents the next generation of ATA interface technology, it departs from ATA with several significant new features and
benefits, summarized below.
Feature Serial ATA Parallel ATA Parallel SCSI
Low-cost implementation
Point-to-point connectivity
Cyclical Redundancy Checking (CRC) on commands (partial)
Hot-plug/hot-swap support
Highly efficient cabling, connectors, backplanes
Table 1. Hard Drive Interface Features and Benefits Comparison
Point-to-Point Connectivity
Point-to-point connectivity is an important feature that provides significant performance and reliability advantages over the shared
connectivity approach employed by both the ATA and SCSI parallel interfaces. Each port on a Serial ATA controller serves just one
-1-
device; that is, the controller communicates with a given drive only through the port where it is connected (see Figure 1
below).
Point-to-Point Loop Arbitrated Technology
ATA/SCSI
ATA/SCSI
RAID
RAID
Controller Serial ATA drive ATA/SCSI
RAID
ATA/SCSI
RAID
Controller
ATA/SCSI
RAID
Serial ATA drive Serial ATA drive
ATA/SCSI ATA/SCSI
RAID RAID
ATA/SCSI ATA/SCSI
Serial ATA drive RAID RAID
Shared Bus
Controller Serial ATA drive Serial ATA drive Serial ATA drive
Figure 1. Point-to-Point Versus Shared Connectivity
Because there is no sharing of the bus, each drive can communicate directly with the system at any time. This means that
the entire available interface bandwidth is dedicated to each device. The dedicated link approach of point-to-point
connectivity eliminates the arbitration delays sometimes associated with shared bus topologies. With a shared bus
approach, this overhead increases as drives are added to the shared bus. So, in a typical ATA or SCSI RAID system, adding
a hard drive will increase the total system throughput by some amount less than the throughput of the disk. With Serial
ATA, on the other hand, each added hard drive can deliver its maximum throughput. Point-to-point connectivity offers
the added benefit of simpler configuration. Dedicated links make a Serial ATA RAID system easy, fast and relatively
inexpensive to set up.
Point-to-point connectivity introduces an important issue relative to scalability. With Serial ATA, the capacity supported
by the configuration is a function of the number of available point-to-point connections. Traditionally, the architecture
would determine the number of devices supported, and the number of connectors on the cable attached would dictate the
number of drives that may be connected. With SATA, scaling is achieved simply by adding more point-to-point links in
the system at the host level, with each connector having one cable and one drive connected to it. The number of links
included is dictated by the number a vendor has included on the system board. Additional links can be added via a
controller card or a RAID card.
Cyclical Redundancy Checking (CRC) Error Detection
Perhaps the most beneficial and most significant improvement of Serial ATA over Parallel ATA is Cyclical Redundancy
Checking (CRC) on commands. While both Serial and Parallel ATA have CRC, the CRC is different in SATA due to the
fact that SATA has CRC for the entire command (FIS - command code, data packets, etc.), where Parallel ATA only has
CRC for the data transfer. CRC has been available in SCSI since Ultra160 technology was introduced, and significantly
improves data integrity by checking all transferred data and verifying that it is received correctly.
CRC greatly increases error detection reliability during the execution of high-speed transfers and hot-plugging or hot-
swapping (insertion and removal of drives without interfering with the data transfers taking place; see
"Hot-Plug Support" on page 3). CRC is calculated on a per-burst basis by both the host and the hard drive, and is stored
in their respective CRC registers. At the end of each burst, the host sends the contents of its CRC register to the hard
drive, which then compares it against its own register's contents. If the hard drive reports errors to the host, then the host
retries the command containing the CRC error.
-2-
With Serial ATA, each protocol layer has the capability to identify errors and can perform recovery and control actions as
well as forward information to the next higher layer in the stack. Each layer has a means to be aware of--and recover
from--errors in the layer below it. A detailed explanation of this feature is covered in "Protocol and Interconnect Level
Reliability" on page 4.
Hot-Plug Support
SATA supports hot-plugging (also known as "hot-swapping"), the ability to swap out a failed hard drive without having to
power down the system or reboot. This capability contributes to both data availability and service-ability without any
associated downtime, making it a critical feature for extending SATA into enterprise applications.
The Serial ATA 1.0 specification requires staggered pins for both the hard drive and drive receptacles. Staggered pins mate
the power signals in the appropriate sequences required for powering up the hot plugged device. These pins are also
specified to handle in excess of the maximum allowed inrush current that occurs during drive insertion. SATA-compliant
devices thus need no further modification to be hot-pluggable and provide the necessary building blocks for a robust hot
plug solution, which typically includes:
s Device detection even with power downed receptacles (typical of server applications)
s Pre-charging resistors to passively limit inrush current during drive insertion
s Hot-plug controllers to actively limit inrush current during drive insertion
s Receptacle and plug guides for alignment during drive insertion
Improved Cabling
Cabling sits near the top of the list of hardware-related service calls. It also is a significant consideration in air flow design
and hardware assembly. Serial ATA features an enhanced cable design that offers important benefits that add up to reduced
service calls and efficient manufacturing.
Though it appears simple, the basic Serial ATA connector design is a remarkably efficient and practical design offering a
number of notable features/benefits:
s The "L" shaped Serial ATA data and power connector make plug orientation very obvious to the end-user, thus pre-
venting incorrect mating.
s The extrusion has "ears" which guide and align the plug during the mating process.
s The conductors are engineered for hot-plugging; they connect in three stages--first pre-charge, then ground, then
power.
s The connector locations on the back of both 2.5" and 3.5" devices are identical, allowing design of backplanes that
accommodate either size device.
The Serial ATA connector represents a substantial improvement over Parallel ATA, which has a long history of problems
with bent pins. And it represents a significant improvement over SCSI's daisy-chain topology, where, if one cable
disconnects or fails or if a terminator is missing, the entire group of drives will not perform properly.
Figure 2. Parallel ATA vs. Serial ATA Cabling
-3-
S E R I AL ATA R E L I A B I L I T Y
Serial ATA features greatly enhance storage system reliability compared to ATA. Unlike its predecessor, SATA provides
important protection and recovery features at four levels: the protocol and interconnect level, the device level, the sub-system,
and the system level.
Protocol and Interconnect Level Reliability
Thanks to its CRC feature (see "Cyclical Redundancy Checking (CRC) Error Detection" on page 2), SATA provides solid
error detection at every layer:
s PHY layer detects and handles raw signal and 10-bit stream problem conditions such as no device, OOB signaling
failures, and PHY internal errors. Information pertaining to these errors is also visible to the link system.
s Link layer detects and handles 10-bit stream and frame problem conditions such as invalid state and data integrity
errors. Information pertaining to these errors is also visible to the transport layer.
s Transport and software layers detect and handle command and status problem conditions such as internal, frame,
protocol, and state errors, including error handling for command block status registers and
command failed/timeout. Information pertaining to the transport layer is visible to the software layer.
As explained previously, SATA features hot-plug ability plus an enhanced cable design that helps eliminate connection
errors and bent pins. Together, these features contribute to improved reliability and reduced downtime resulting in fewer
service calls.
Device Level Reliability
Hard drive reliability is expressed in number of hours as Mean Time Between Failure (MTBF). It is important to note that
Western Digital's first SATA drive will have the same MTBF (1.2 million hours) as a SCSI drive. Keeping this in mind,
also consider that the statistical mean represented by MTBF provides information about a population, but not particular
drives. The failure of a particular drive cannot be predicted--hard drive failure is an ever-present reality. Therefore, it is
reasonable to expect the reliability of SATA hard drives to approximate that of SCSI drives.
To address the reality of potential hard drive failure, storage vendors utilize RAID (Redundant Array of Inexpensive Disks)
to provide disk mirroring and parity data protection. In the 20 years since RAID was created, the "inexpensive" in the
acronym has evolved to "independent," reflecting the evolution toward more expensive drives to achieve higher RAID
performance.
Serial ATA offers price and performance that puts the "inexpensive" back in RAID. In addition to its low cost, Serial ATA
has characteristics that make it particularly good for RAID including dedicated point-to-point channels with a
manageable cabling topology and staggered spin-up. The result is an affordable storage solution that delivers more than
satisfactory uptime for a range of applications not considered "mission-critical."
Sub-System Level Reliability
The physical hardware of a storage system also impacts its availability. The enterprise storage market is typified by storage
arrays that provide dense, rack-optimized solutions with redundant components such as power supplies and fans. Systems
manufacturers are striving for ever-denser solutions, requiring smaller form factors, to meet customer demand for more
storage in limited space. This means the number of drives requiring cooling is increasing. This growing challenge for
storage enclosure manufacturers is exaggerated when utilizing ultra high-density SATA drives.
Temperature is a critical factor impacting the reliability of drives and other components--the cooler the drives, the greater
the reliability. As enclosure manufacturers develop solutions based on these ultra high-density drives, it is critical to ensure
proper cooling and fan redundancy to preserve system availability. SATA is well-positioned to handle these environments
since its thinner cabling allows better airflow through the system.
-4-
System Level Reliability
To achieve high system availability (as well as scalability), many solution providers employ cluster configurations. There
are two basic cluster configuration models: the shared disk model and the shared nothing model. Serial ATA can be
employed effectively in either of these two models.
Shared Disk Cluster Model Shared Nothing Disk Cluster Model
Client Client Client Client Client Client Client Client Client Client
LAN LAN
Servers
Servers
Storage Array w/
Serial ATA Drives
Storage Array w/
Serial ATA Drives
Storage Storage
Controller Controller
Fibre Channel,
Fibre Channel,
SCSI, etc.
SCSI, etc.
Serial ATA
Serial ATA
NFS or CIFS over Ethernet Serial ATA
Fibre Channel, SCSI, etc.
Figure 3. Cluster Configuration Models
Both clustering models eliminate the server (or appliance) as single points of failure. And, in both cases, Serial ATA RAID
arrays can eliminate the hard drive as single point of failure.
Summary--Serial ATA Reliability
Based strictly on MTBF specifications, SCSI drives clearly deliver a higher level of uptime than SATA drives. However, it
is system uptime that really matters. Since no hard drive is infallible, the only way to meet the need for 24/7 availability is
to create a system that can tolerate a failure of any component, without loss of connectivity. The only way to achieve this
is to create a system that has redundant everything--from multiple I/O controllers and servers to multi-path cabling.
Dual port SCSI drives can add value to such a redundant solution; if it is in a RAID volume, no data or connectivity is lost
if the drive itself fails. However, a single port drive can achieve the same objective through RAID mirroring, such as RAID
5+1 (this is striping with parity + mirroring). Although mirroring requires twice as many drives, many IT professionals
intentionally implement this method for enhanced performance, offering the ability to access the same data in parallel
from both drives (~2x performance).
Additional technologies are also used to aid IT professionals in the ability to predict failures and to quickly swap out faulty
components. System solutions include such implementations as rigid backup policies, hot spares, hot-plug, enclosure
management/error prediction (SMART), or any combination of these system solutions and others as determined by the
needs of each business.
Thus, if one can achieve the system goal of guaranteed data reliability and data availability with either SCSI or SATA
technology in a RAID array, then the deciding factor will likely be cost. In such a scenario, SATA offers an economic
advantage that is hard to ignore.
S E R I AL ATA P E RF OR MA N CE
Since devices based on SATA are still in development, the performance of these devices has yet to be fully tested. However,
it is clear that SATA will usher in an entirely new level of ATA performance, utilizing a data transfer rate of 1.5 Gb/second
(150 MB/s). This surpasses today's mainstream ATA performance by 50% or more, with a clear roadmap for future
performance increases. Recall that in a system combining multiple devices, such as a RAID configuration, SATA offers the
advantage of additive device performance, devoting full bandwidth to each drive in the system.
-5-
A typical misconception about SATA is that its serial nature makes it inherently slower than parallel bus designs. However,
witness the trend toward serial technology for all desktop/server data transport mechanisms including PCI Express, Fibre
Channel SANs, Infiniband Architecture, and others. Indeed, even SCSI is moving from parallel to serial technology, with
the future Serial Attached SCSI (SAS) draft standard. Given the high speeds required by today's technology, coupled with
the synchronization constraints of parallel data buses, high-speed serial links provide a practical basis for future
technologies.
How SATA Improves Performance
The key to SATA's higher performance (at least compared to ATA) is its point-to-point topology. SATA does not have to
share the ATA bus as in the traditional ATA master/slave topology. Add to that SATA's dedicated 1.5 Gb/second
(150 MB/s) maximum performance per device, and it can be seen that the bus already has room to spare when today's
best-of-breed drives are hard pressed to deliver 100 MB per second. Note that this is just the starting point for Serial ATA;
the specification developers are planning increased speed transitions for SATA over the next several years.
Performance and Price
While it is likely that first-generation SATA drives may not match the throughput of the best high-end SCSI drives, it is
important to place this in the context of real-world storage solutions. Few storage solution vendors actually use high cost,
best-of-breed SCSI devices in their systems. The SATA cost advantage enables vendors to use best-of-breed SATA products
with adequate performance at a cost lower than if they used "mainstream" SCSI devices. This may in reality deliver
performance comparable to or less than the best SATA units. The important point about SATA is that it narrows the gap
between ATA and SCSI performance, while retaining the traditional ATA price advantage. And, in today's IT
marketplace, adequate performance at the best price point is the name of the game for a growing number of storage
applications.
C O N CL US I ON
Clearly, SATA represents an important extension in proven ATA interface technology--one that targets SATA squarely at
networked storage. The question is not if SATA will penetrate the enterprise, but how far can it go? While its attractive
cost has already driven ATA technology into the enterprise space, SATA delivers performance and reliability that promise
to solidify ATA's presence in the entry level/blade server category and extend to other business critical systems. SCSI will
likely continue to be used for mission-critical storage applications. SATA, however, is ideally positioned to satisfy the
requirements for a large number of enterprise storage applications, including:
s Web hosting
s Firewalls
s E-mail servers
s Small business and remote office business applications
s File sharing
s Media streaming
s Tape backup replacement
s Performance workstations
From a system point of view, SCSI and Fibre Channel will likely continue in high-end Storage Area Network (SAN)
segment. However, despite the growth of SANs, Direct Attach Storage (DAS) still dominates the storage market. These
DAS configurations are attractive to organizations seeking moderate performance at a reasonable cost. This is a market
space where Serial ATA-based storage solutions can provide a superior value proposition, appealing strongly to what is still
the largest segment of the external storage market.
DAS customers can take advantage of new and attractive options that Serial ATA offers. Now they can choose enterprise-
class storage subsystems with enhanced power, packaging, acoustic, and cooling characteristics within a fully redundant/
fault-tolerant configuration; a feature set superior to existing SCSI JBOD arrays. In addition, the use of low-cost Serial
-6-
ATA external RAID controllers within such subsystems in combination with Serial ATA drives will provide a new entry
point in terms of cost per gigabyte in the RAID storage market space.
Storage solutions with higher performance and higher price tags than ATA technology will continue to receive serious
consideration in the storage industry. The question for storage solution providers and customers is this: Is the information
of such critical importance that it warrants paying the significant premium SCSI- and Fibre Channel-based systems will
command over a SATA based solution?
Given today's emphasis on IT costs, the exploding growth of storage capacity and the attractive performance roadmap and
reliability features offered by Serial ATA, it is likely that many IT managers, especially in the Small and Medium Enterprise
(SME) space, will say "Yes" to SATA for a growing number of enterprise storage applications.
F R E Q UE N T L Y A SK E D Q U E ST I ON S ( FA Q S )
Is Serial ATA backward compatible with Parallel ATA?
Serial ATA is specified for software compatibility with Parallel ATA. In other words, no changes are required to operating
systems and PATA drivers to use SATA hard drives. Computer vendors who would like to incorporate the advantages of
Serial ATA will provide adapters that enable backward compatibility for SATA hard drives in today's computer systems.
Why is Serial ATA an acceptable alternative to SCSI for enterprise storage?
Serial ATA meets the demanding requirements of server and networked storage systems; needs that go beyond those of
desktop systems. These critical requirements include:
s Performance