Вы находитесь на странице: 1из 11

Disk drive technology overview

technology brief

Abstract.............................................................................................................................................. 2
Introduction......................................................................................................................................... 2
Classes of drives.................................................................................................................................. 2
Housing.............................................................................................................................................. 3
Single drive capacity, performance, and reliability .................................................................................. 3
Single drive performance .................................................................................................................. 3
Single drive reliability....................................................................................................................... 5
The HP disk drive qualification process ............................................................................................... 6
New technology .............................................................................................................................. 6
Using multiple disks to improve performance and reliability ...................................................................... 7
Connecting the physical disk to the system.............................................................................................. 7
ATAAdvanced Technology Attachment ............................................................................................ 7
SCSISmall Computer System Interconnect ........................................................................................ 8
SASSerial Attached SCSI ............................................................................................................... 8
SATASerial Attached ATA.............................................................................................................. 9
Comparison and discussion............................................................................................................... 9
Summary .......................................................................................................................................... 10
For more information.......................................................................................................................... 11
Call to action .................................................................................................................................... 11
Abstract
This paper reviews the classes of disk drives, the key factors determining capacity, performance and
reliability in single drives, the use of multiple drives to further increase performance and reliability,
and the options available to connect the drives to the system.

Disk capacity has increased at rates similar to those of microprocessor performance. New and
improved interconnect technologies allow the rapid transfer of large amounts of data to and from the
disk. New and more cost-effective applications are made possible by increased storage capacity and
reduced cost per bit.

HP innovation and engineering deliver industry-leading disk technologies that optimize overall system
capacity, performance, reliability, and value.

Introduction
Disk drives provide the persistent secondary memory in almost all servers, desktop computers, and
notebook computers. Disk drives are also increasingly common in portable electronic devices such as
music players and automobile navigation systems. The global market is about one million units per
day.

Other recording technologies (solid-state flash memory, optical disk, and magnetic tape) are also
used for secondary storage, but the remainder of this technology brief will focus on disk drives with
rotating magnetic media.

The key differences between main memory (semiconductor RAM) and secondary memory (magnetic
disk drives) are access speed and capacity. Accessing secondary memory is typically much slower
than accessing main memory (approximately 1000 times). Secondary memory is typically much
larger than main memory (approximately 100 times or more).

Most of the engineering in disk drive and interconnect technologies is driven by a desire to reduce
this difference in access speed while simultaneously increasing disk drive capacity and reliability.

Innovative strategies to increase capacity, performance, and reliability have resulted in dramatic
increases in disk capacity and performance.

Classes of drives
Disk drives may be placed into three classesConsumer, Nearline, and Enterpriseaccording to their
intended application.

The intended application heavily influences the design and selection of components for the disk,
resulting in a good-better-best hierarchy.

Consumer-class drives are designed to achieve maximum capacity per dollar. Typical intended
applications are bulk/archival storage, standard desktop systems, and standard notebook systems
where the I/O (Input/Output) workloads are limited and desktop/notebook reliability is acceptable.
These drives may be acceptable in some non-mission-critical server applications where cost is most
important and the customer is willing to trade off performance and reliability.
Nearline-class drives are designed for high reliability and desktop performance. Typical applications
are bulk storage, where I/O workloads should be limited to achieve acceptable reliability.

Enterprise-class drives are designed for maximum reliability, maximum performance, scalability, and
error management under heavy 24x365 I/O workloads. They are intended for mission-critical
applications. Enterprise-class drives are the only class of drives intended for unlimited I/O loads.

Housing
HP provides two lines of Universal Carrier for disk drives, one for small form factor (2.5 inch) drives
and one for large form factor (3.5 inch) drives, which allows any hot-pluggable drive from a family to
fit mechanically and electrically with any industry standard server or storage product.
This mechanical commonality extends to all SAS-based StorageWorks and HP Integrity server
products, as well. This provides unparalleled customer value, allowing mixed HP Enterprise solutions
to be supported by a common family of hard drives.

Single drive capacity, performance, and reliability


The capacity of a disk drive is determined by the number of platters, the surface area of each platter,
and the number of bits that can be stored per unit area. Capacity is set at manufacturing.

The number of bits that can be stored per unit area is called areal density. Areal density is
determined by the number of tracks-per-inch of disk radius multiplied by the number of bits--per-inch of
track.

Disk drive capacity is measured in gigabytesusually hundreds of gigabytes in todays drives.

A common source of confusion regarding disk drive capacity is the definition of a gigabyte. In a disk
drive a gigabyte is exactly 1,000,000,000 bytes, but operating systems often use the binary-based
approximation of 230, or 1,073,741,824 bytes, per gigabyte. Thus, a disk drive with 100 actual
gigabytes of storage may be reported by the operating system as having only 93 gigabytes.

Single drive performance


A number of factors determine the actual performance of a disk drive. These include the rotational
speed of the platters, seek performance, mechanical latency, read/write bandwidth, queuing
strategies, and interface technologies.

In the ideal case, the head is positioned at the target segment above the correct track when the read
or write request is received, and data transfer can begin immediately. In this ideal case, the
rotational speed of the platter is the controlling factor.

If the rate of rotation of the disk platter can be doubled, data can be read from or written to the track
twice as fast.

In the non-ideal case, the head must be moved above the correct track and then must wait for the
target segment to pass under the head. This mechanical latencythe time to move the head to the
correct track and then wait for the target segmentis called the seek delay or seek time.
Seek delay is fundamental to disk system performance. Seek time is measured in milliseconds, and
typical values are 4-10 ms. A number of strategies have been developed to directly or indirectly
avoid or reduce this mechanical latency, and are discussed further below.

Disk drive performance is usually characterized under two data transfer scenarioscontinuous data
transfer rate off of the media, and random IOPs (Input/Output operations per second).

Continuous data transfer occurs when reading or writing relatively large blocks of data to sequential
disk sectors, and sets the upper bound of performance for the drive. It should be noted that the
maximum continuous data rate is valid only for the outermost tracks on the drive, and that this rate
can be up to 50 percent lower on the inner tracks.

Random access occurs when reading or writing relatively small blocks of data to sectors which may
be scattered across the disk. The speed of the actuator bounds performance in this scenario, and sets
the lower bound of performance for the drive.

Actual real-world performance is heavily dependent on the nature of the application dealing with
large blocks of sequential data (for example, video files) versus small blocks of unrelated data (for
example, customer records in an e-commerce database).

As the disk fills up, large blocks of data may have to be written to non-sequential segments or non-
adjacent tracks. This scattering of data across the disk is called fragmentation, and can
significantly degrade performance.
Table 1. Strategies to improve single disk capacity, performance, and reliability

Mechanical Magnetic Disk I/O Other


increase platter increased bit density write cache buffer bad sectors retired
rotation rate per unit of track data to be written to
disk

reduce distance queue read spare sectors


between adjacent operations
tracks

reduce platter size queue write error correcting code


operations

decrease seek times reorder read and sealed unit


write operations to
execute the next
operation physically
available on drive

Mechanical engineering strategies reduce the physical distance that the read/write heads must travel
to reach the target segment. These include:

smaller diameter platters


multiple platters per drive
increasing the speed of rotation of the platter
reducing the distance between tracks
decreasing the seek time to a track

Magnetic storage strategies increase the amount of data in each track by increasing the bit density
per unit length of track thus reducing the need to move among tracks and allowing immediate read-
after-write verification.
Disk I/O strategies reduce the time that a logical read/write spends waiting for the physical
read/write operation. These optimizations seek to effectively decouple the logical and physical
operations of the disk. Increasingly sophisticated approaches become practical as the embedded
processing power and memory incorporated into the drive increase. These approaches include:

buffer the data to be written to disk (write cache)


queue read operations
read-ahead caching
queue write operations
write-back caching
reorder read and write operations to execute the next operation physically available on the
drive

Single drive reliability

Disk drive reliability is measured in terms of AFR (Annual Failure Rates). This is the percentage of
disk drive failures in a large population of drives in operation for one year. For example, 100,000
drives with an AFR of 1.5 percent would experience approximately 1,500 failures per year. AFR
calculated from a small number of drives is subject to large statistical variations rendering it
inaccurate.

A major factor in determining reliability in actual use, and the source of much confusion, is the Duty
Cycle and I/O Workload to which the drives are subject. Duty Cycle is simply power-on time,
which is calculated as the number of hours that the disk drive is powered on divided by the
number of calendar hours. HP assumes a 100 percent Duty Cycle for all drives. I/O Workload is
disk working time, which is calculated as the number of hours that the disk drive is aggressively
reading and writing data divided by the number of calendar hours.

Enterprise drives are designed for unlimited I/O workloads, that is, for continuous I/O activity.
Nearline and Consumer Drives are designed for IO workloads of 40 percent or less.

If any doubt exists about the expected workload, and if reliability is a priority, then Enterprise class
drives should be selected.

Drives are subject to mechanical problems created by shock, vibration, environmental extremes, and
thermal effects. These problems may degrade performance or reliabilityfor example by causing
the read head to be displaced from the data trackor cause the loss of data or even catastrophic
failure of the drive. Enterprise drives are the most robust to vibration effects, followed by Nearline
drives that have a lower tolerance to vibration. Desktop and notebook drives are not designed to
perform in high vibration environments and will exhibit degraded performance. Humidity, corrosive
atmosphere, and static electricity may also cause disk performance degradation.

Temperature is a key factor influencing reliability and is usually best managed by controlling the
operating environment.

Multiple drives in a single enclosure may interact to create coupled vibration problems. This problem
can result from using Consumer or Nearline class drives with Enterprise class I/O workloads.
The HP disk drive qualification process

HP employs rigorous qualification and quality processes to ensure that the disk drives it ships remain
reliable, meet customers requirements, and integrate seamlessly into HP server and storage systems.
In addition, the HP process assures customers that continuous improvement occurs in both current and
future products and processes. These actions benefit the supplier, HP, and most importantly, the
customers. The qualification process consists of four specific steps:
1. Selection Evaluation
2. Development Verification
3. Supplier Production Qualification
4. Continuous Improvement / Performance Monitoring
HP development engineering works closely with the disk drive suppliers to execute a comprehensive
set of approximately 50 different procedures and specifications that determine the testing and metrics
that a candidate drive design must satisfy. Typically, approximately 1000 unique hard disk drives are
used to evaluate a product family during the Selection Evaluation and Development Verification steps,
and approximately 2 million drive test hours transpire.
The Supplier Production Qualification phase includes a thorough analysis of the suppliers
capabilities, focused on validating supplier process capability and process controls, and measured
product quality.
The analysis includes extensive review of the suppliers process controls, closed-loop corrective action
processes, and overall quality control system. The final stage of the Supplier Production Qualification
includes a comprehensive analysis of the products quality performance via the HP configuration pilot.
Disk drive products that pass the extensive HP qualification process and get released then proceed
into HPs Disk Drive Performance MonitoringContinuous Improvement phase during volume
production. Three main areas of focus exist:
Validate volume production is in process control
Measure, analyze, and react to product quality data
Deliver continuous product improvements
HP and the disk drive suppliers work closely together as a team during the volume production phase
of a product. The team monitors the performance of each product through quality control methods at
both the suppliers factory and HP option kitting configuration sites. Product quality data is reviewed
on a daily, weekly, and monthly basis.
For further details, see the white paper at
http://h20000.www2.hp.com/bc/docs/support/SupportManual/lpn12758/lpn12758.pdf.

New technology
The major trends in disk drive technology are smaller form factor and increased density. For
example, 2.5 inch drives require half the power and generate 70 percent less heat than 3.5 inch
drives.
Using multiple disks to improve performance and reliability
Drives fail. It is a fact of lifetherefore storing data on a single disk drive presents an exposure to
data loss. It is recommended to always use some form of fault-tolerant RAID (Redundant Array of
Inexpensive Disks) across multiple drives.

RAID strategies can be characterized by how data reliability is achieved (how parity or other error
correction data is distributed across the array), the minimum number of drives required, and data
storage efficiency. The performance of many drives is better than that of one drive.

The choice of RAID strategy, and the particulars of the implementation, impact read performance,
write performance, and robustness in the face of hardware failures. Both read and write performance
also varies by workload, that is, whether I/O (many small data units) or bandwidth (fewer, large data
units) predominate.

Raid 0 striping to two or more disks; no redundancy, performance improvement only


Raid 1 mirroring; duplicates same data on two disks; redundancy and potential performance
improvements
Raid 1 + 0 mirroring and striping; redundancy and performance improvement
Raid 5 block striping with distributed parity; three or more drives; fault tolerance
Raid 6 block level striping with dual distributed parity; three or more drives, increased fault
tolerance

Advanced Controllers

Advanced controllers (such as the HP Smart Array) decouple the logical disks seen by applications
from the physical devices used to implement the disk subsystem. A single logical disk (as seen by an
application) may be mapped onto an array of multiple physical disks. These controllers include both
hardware and software.

This approach provides greatly enhanced flexibility, expandability, maintainability, and performance.
Smart Array controllers are available for SAS, SATA, and SCSI interfaces.

For further details, see the technology brief entitled HP Smart Array Controller Technology, at
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00687518/c00687518.pdf.

Connecting the physical disk to the system


Various interconnect technologies are used to connect one or more disk drives to the computer system.
In servers, the most common choice is Serial Attached SCSI (SAS). See Table 2 below for a
comparison of key features. Older technology interfaces ATA and SCSI are included for reference.

ATAAdvanced Technology Attachment

ATA is often used interchangeably with IDE (Integrated Drive Electronics). IDE denoted that the
controller was in the drive enclosure, rather than an external card. Enhanced IDE expanded the
maximum drive capacity from 528 MB to 8.4 GB. The terms ATE, IDE, EIDE are often used
interchangeably. The ATA interface is no longer recommended for new installations. It is included
here because of its widespread legacy deployment, particularly in lower-end servers and desktops.

ATA uses a parallel interface, with a 40-pin, 80-conductor ribbon cable. The cable is bulky, and
impedes air flow. A two-device controller is usually built into the motherboard. Transmission rates
(133 MB/sec) have reached the maximum that can be achieved with the electrical design of the
interface. Data transmitted between the controller and disk is CRC encoded, but commands are not.
Hot swapping of disks is not supported.

SCSISmall Computer System Interconnect

SCSI may be considered for use where good performance plus high capacity is required, for
example, in some workstation applications.

SCSI improved on ATA, offering better performance, support for more devices, and a broader
command set.

SCSI 320 Ultra is the seventh generation SCSI implementation, and represents the technical and
economic limits of the original parallel SCSI architecture. SCSI uses a parallel interface, with a 68
pin cable (flat ribbon inside the server, round outside). The controller may be built into the
motherboard or provided as an adapter card. Up to 16 devices, including the controller, are
supported, and share a single common bus. Transmission rates up to 320 MB/sec are supported.
Hot swapping is supported.

SCSI is being replaced by SAS. For further details on SCSI, see the Technical Brief entitled Ultra320
SCSI technology in HP Smart Array Controllers at
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00687518/c00687518.pdf.

SASSerial Attached SCSI

SAS is usually considered the most cost-effective solution for mission critical, high I/O workload
applications, such as business critical databases.

SAS uses a point-to-point serial interface, with each device connecting directly to a SAS point. The
serial interface allows the complete link bandwidth available to each device, greatly increasing
performance and scalability. In addition, the links are full duplex, and can be grouped to further
increase bandwidth. SAS inherits the proven SCSI command set.

SAS disk drives are dual ported, providing for two active-active paths to each device. Please see the
section entitled, For more information, for other papers and resources.

SAS built upon the SATA physical characteristics. This means that SATA drives can be used with SAS
controllersin fact, SATA and SAS drives can be mixed in a single enclosure (SAS devices cannot be
used with SATA controllers).

First generation SAS supports 3 Gb/sec (about 300 MB/sec). The second generation provides
6 Gb/sec.
HP was instrumental in developing the SAS standard. For a more detailed discussion of SAS, please
see the Technical Brief entitled, Serial Attached SCSI technology at
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00302340/c00302340.pdf.

SATASerial Attached ATA

SATA is usually considered the best solution for price-sensitive low-I/O workload applications, such as
near line storage, and is also expected to dominate the desktop market due to low cost.

SATA introduces a serial communication interface that operates in simplex mode, increases data
transfer rate, requires a small-diameter cable, supports additional disks, and supports hot swapping.

SATA uses a 7-pin small diameter cable. The controller may be built into the mother board or
provided as an adaptor card.

The SATA specifications was initially released in three variants:


1.5 Gb/sec
1.5 Gb/sec with extensions
3.0 Gb/sec

The initial SATA 1.5 Gb/sec variant was targeted at replacing ATA in the desktop and consumer
markets. It introduced a serial interface that supports 1 drive per controller port.

SATA 1.5 Gb/sec with extensions is targeted to workstations and low-end servers, and adds native
command queuing.

SATA 3.0 Gb/sec is targeted to workstations and low-end servers, and increases the data transfer
rate.

The SATA Roadmap calls for SATA to eventually reach speeds of 6 Gb/sec.

For a more detailed discussion of SAS, please see the Technical Brief entitled, Serial ATA
technology at
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00301688/c00301688.pdf

Comparison and discussion

Certain capabilities have traditionally been inherent in SAS or SATA, but this is changing. The
benefits and constraints of these two interfaces may be blurred over the next year or two.

Interconnect bandwidths are now exceeding the bandwidth available from the physical drives, with
SAS moving to 6 GB/sec, and SATA to 3 GB/sec. Disk drives will not be able to fully use these
bandwidths until well beyond 2012.

HP SAS Smart Array Controllers, coupled with Enterprise class drives, provide the best performance
and highest reliability, and are recommended for mission-critical industry-standard server solutions.
Other technologies may be appropriate for non-mission-critical applications.
Table 2. Comparison of major interfaces for Industry-Standard Servers

SCSI Ultra320 SATA 1.5e SAS

Typical Use workstation near line mission critical


low end server low IO workload high IO workload
mission critical best performance
good performance +
capacity

Capacity, typical 72 - 300 60 - 750 36 - 300


drive (GB)

Seek time, typical 3.8 5.4 9 - 12 3.5 4.0


drive (ms)

Architecture shared point-to-point point-to-point


parallel bus serial bus full duplex
16 bit serial bus
Maximum Throughput 320 MB/sec 150 MB/sec 300 MB/sec

Cable Length 12 m 1m 8m

Devices 16 15 1 16,256

Command set SCSI ATA SCSI +

Hot swap yes yes yes

Drive ID user configured worldwide unique ID


set at factory

Drives supported SCSI SATA SAS or SATA


only only can be mixed

Summary
Multiple drive configurations (RAID) provide increased reliability and can provide significantly
increased performance.

Enterprise class drives with SAS interconnects provide the highest performance and most reliable
storage for industry-standard servers.

Disk technology is evolving rapidly. Hardware should be reviewed periodically to ensure that the
most cost-effective solutions are deployed.

1
The total number of devices when SATA drives are used in a SATA infrastructure. This entry does not address using SATA drives in a SAS
infrastructure.
For more information
For additional information, refer to the resources listed below.
HP Hard Disk Drive Quality System
ftp://ftp.compaq.com/pub/products/servers/proliantstorage/drives-enclosures/hdd-quality.pdf

Serial ATA Technology


http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00301688/c00301688.pdf

Serial Attached SCSI Technology


http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00302340/c00302340.pdf

Smart Array Controllers


http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00687518/c00687518.pdf

HP Disk Drives for ProLiant Servers


http://h18004.www1.hp.com/products/servers/proliantstorage/drives-enclosures/index.html

Serial ATA Organization


http://www.sata-io.org/

Call to action
Send comments about this paper to TechCom@HP.com.

2007 Hewlett-Packard Development Company, L.P. The information contained


herein is subject to change without notice. The only warranties for HP products and
services are set forth in the express warranty statements accompanying such
products and services. Nothing herein should be construed as constituting an
additional warranty. HP shall not be liable for technical or editorial errors or
omissions contained herein.

TC070505TB, May 2007