2. Abstract
Introduction
Overview of disk drives
Performance overview
Disk controller
Reliability
Energy consumption overview
Design constraints
New disk drive architectures
New storage media and storage devices
Discussions
Conclusions
3. This article explain:
Importance of energy efficiency in designing
disk drive storage systems.
Hard drive design have reached a turning
point at which they have to be reborn in order
to maintain high reliability and energy
efficiency.
The evaluation of disk drive over 5 decads.
4. Disk drive is
◦ most important storage , offers high performance
◦ large capacity ,high reliability
Since IBM 1301 disk drive was announced in 1961, disk drives
have experienced dramatic development to meet capacity,
performance, and other capability requirements.
The 1301 stored 28 million characters on a single module
IBM 1301 disk drive
5. Magnetic recording technology
Has two mile stones
Longitudinal recording and Perpendicular recording.
Hard disk technology with longitudinal
recording has an estimated limit
of 100 to 200 gigabit per square inch.
Perpendicular recording was used
in 2005 for hard disks.
Perpendicular allow information densities of up to
around 1 Tbit/sq. inch (1000 Gbit/sq. inch).
In August 2010 drives with
densities of 667Gb/in2 were available commercially.
6. Traditional longitudinal recording technology has
successfully achieved 100% growth of areal density (AD)
AD: is a measure of the number of bits that can be
stored in a unit of area. It is usually expressed in
bits per square inch (BPSI).
Areal density is computed as the product of two
density measures.
◦ Track Density(TPI): How many tracks can be placed
down in inch of radius on the platters.
◦ Linear or Recording Density(BPI): How tightly the bits
are packed within a length of track. If in a given inch
of a track we can record 200,000 bits of information,
then the linear density for that track is 200,000 bits
per inch per track (BPI).
7. Superparamagnetic effect (shrinking volume of
magnetic grains) poses a serious challenge for
further increases of the AD. The reason is that
each bit cell in a track is composed of multiple
magnetic grains.
Bit-cell composed
of magnetic grains
50-100 grains/bit
8. But, the grain size cannot be decreased much below a
diameter of ten nanometers. Using fewer magnetic grains in a
bit cell requires more complicated error correcting codes.
The growth mainly depends on the improvement of
Revolutions Per Minute (RPM), magnetic recording technology,
the size of the on-board cache, along with some reductions
in the seek time [Hitachi 2009].
Trends of recent years imply that flash memory could be a
good candidate for bridging the performance gap.
However flash-based storage devices are still expensive, and
their characteristics such as write endurance (each memory
cell has a limited number of times that it can be erased
before the memory cell fails ) and erase before write are still
challenging problems.
9. Hard disks mainly consists of platters, spindle, disk arm, disk head,
motor, controller, etc.
The platters spin at a constant rate called RPM. The data is recorded
magnetically in concentric tracks on the platters.
10. Disk access time:
Taccess = Tseek + Trotate + Ttransfer
◦ Seek Time. Time it takes to locate a particular piece of
information .
◦ Rotational Latency: when the disk head arrives at the target
track, it must wait for first sectore before it begins to transfer
data.
◦ Data Transfer Time. Data transfer time is the amount of data
divided by the data transfer rate.
11. Data Transfer Time
◦ Consists of two parts
External data transfer: Rate between memory and
disk cache
Internal data transfer :Rate between disk cache and
disk storage media
Internal data transfer is also called Internal Data Rate
(IDR).
Expressed in MB/s
IDR is much lower than the external data rate because
reading or writing on disk plate is time consuming.
12. Geometric features:
◦ Outer tracks on disk platters are much larger than the inner
tracks. Modern disk drives employ a technique called ZBR.
◦ Zone Bit Recording (ZBR) is used by disk drives to store
more sectors per track on outer tracks than on inner tracks.
◦ ZBR results in a much smaller data transfer time of outer
zones than that of inner zones.
13. A disk controller contains
Storage interface offers a standard protocol
(e.g., IDE, SCSI, FC, SATA, etc.)
Disk sequencer
Manages the data transfer between storage
interface and data buffer.
ECC
Responsible for error checking.
Servo control
Detects the current position of the disk head
Microprocessor
Overall control of disk drive.
Disk cache
Temporary storage.
14. Disk cache
Temporarily holding data
Principles of data locality to improve hit ratio
All modern disk drives contain a small amount of on-board cache
(RAM) to speed up access to data on a disk drive.
Cache replacement algorithms(Already done in OS)
Random Replacement(RR)
RR replaces cache lines by randomly selecting a cache line
very fast,
requires no extra storage
easiest one to implement
performs poorly(a page that will used in near future may be swapped)
Least Frequently Used (LFU)
Which have been used least frequently are evicted.
Recently active but currently cold cache lines
Increases the miss ratio and reduces the cache performance.
Least Recently Used (LRU)
◦ Evicts those cache lines used least in the recent past on the assumption that they will not be used in
the near future
15. Disk Scheduler
Queue the incoming requests.
Scheduling algorithms
First Come First Served (FCFS)
Shortest Seek Time First (SSTF)
SCAN algorithm
Reliability
Disk drive consists of one or more platters rotating on a
common spindle
Spindle motor is adopted to spin the platters and
maintain the RPM.
They may fail due to various component failures (e.g., disk
head, media, firmware, etc.)
The environmental factors, including temperature, humidity,
vibration, etc all have impacts on the failure of disk drives
16. Spin down = Idle to standby
Spin up = standby to active
Table I. The Major Characteristics of Five Different Disk Drives
17. Energy Conservation Methods
◦ 1. Timeout strategy
Once a disk drive is idle ,the disk is spun
down in an effort to save energy. Widely used.
2. Application-aware power management
Requires modifying existing applications,
which makes it impractical.
3. Compiler-driven method for disk power
management has been suggested .
4. Bucket method : Extending the idle length so
energy is be conserved.
18. Performance of disk drives has been
experiencing 40% growth per year, a number
of constraints pose challenges to continue
the 40% growth rate.
The growth of AD results in decreased seek
time and increased IDR.
The perpendicular recording technology will
also reach its limits soon, and new
technologies will be required [Perpendicular
Recording2009].
19. Disk cache can be improve the performance of disk drives by
avoiding slow mechanical latency(measured in milliseconds,
include both seek time and rotational latency).
Accessing a byte of data in cache is much faster than
accessing a byte on the rotating magnetic disk media.
However, studies have indicated that if the disk cache size
grows beyond its limits may cause performance penalty.
Heat generated by certain actions within the disk drive
which effect reliable operation.
High temperature = Head crashes
20. Increasing the RPM can improve disk drive
performance significantly.
Unfortunately, disk drives rotating at speeds
exceeding 20, 000 RPM have been researched but
not commercialized due to heat generation, power
consumption, noise, vibration.
Therefore, it is a big challenge to design new disk
drive architecture which could further advance disk
performance.
21. Power states transition is not applicable to the server disk drives.
Dynamic Rotations Per Minute (DRPM)is proposed for power
management in server disk arrays.
The DRPM technique dynamically modulates the rotational speed
of disk drives so that the disk can serve requests at different
RPMs.
Multispeed disk approach is also suggested to conserve energy.
EED [Deng et al. 2008b] is an energy-efficient disk drive
architecture by extending the length of idle intervals .
A disk drive including two or more spindles each carrying one or
more platters was introduced. Hard Disk Drive with Multiple
Spindles [2011].
22. Flash Memory
Nonvolatile,electrically erased and reprogrammed
small physical size ,lower power consumption
high performance ,Used in digital cameras, MP3 players,
mobile phones, etc.
Two major types of flash memory
◦ NOR: Byte accessible, mainly used for EEPROM
replacement
◦ NAND: Block accessible ,faster erasing and write times,
higher data density. Writing is done in a unit of one page
, erasing is done in blocks.
◦ NAND flash memory can play two roles
As an extension to RAM, and a layer between RAM and
traditional disk drives.
Replacing traditional disk drives as a new block storage
media.
23.
24. Promising Storage Media
◦ Flash memory
◦ Magnetic Random Access Memory (MRAM)
Combines a magnetic device with standard silicon-
based microelectronics to obtain the combined
attributes of nonvolatility, high performance, fast
programming.
◦ MicroElectroMechanical System (MEMS)
MEMS-based storage is a nonvolatile storage
technology that merges magnetic recording material with
thousands of probe-based recording heads to provide
online storage [Schlosser et al. 2000].
However, both MRAM and MEMS are still in
their infant phase of development.
25. Phase-change Random Access Memory
(PRAM)
◦ Mature technology than MRAM or MEMS.
◦ Samsung introduced a 512Mb PRAM
◦ Expected to be the main memory device and to
replace the high density NOR flash within the
next decade [Samsung 2009d].
PRAM can rewrite data without having to first erase data
previously accumulated, it is effectively 30 times faster than
conventional flash memory.
Hybrid Disk
◦ Performance Gap within Disk Drives
Disk drives normally use conventional main memory
(SDRAM) as disk cache.
SDRAM has access time ranging from7–10
nanoseconds.
26. Anatomy of Hybrid Disk.
◦ DRAM are nanosecond devices.
◦ NAND chips are microsecond devices.
◦ NAND flash memory can play as an intermediate layer (e.g., a
nonvolatile cache) between the DRAM and traditional disk
drives.
◦ Hybrid disk integrates NAND flash memory into a standard
disk drive as a second level cache.
◦ Hybrid disk consists of three layers: disk cache, NANDflash
memory, and magnetic platters.
Solid State Disk
◦ SSD refers to semiconductor devices
◦ consists of either DRAM volatile memory or NAND
flash nonvolatile memory.
◦ DRAM-based SSD requires an internal battery and
backup disk drive.
◦ current SSDs employ nonvolatile flash memory as the
storage media (e.g., USB memory sticks).
27.
28. The challenges (e.g., endurance cycles, erase before write,
etc.) presented by NAND flash memory indicate that the flash
memory could fail before the magnetic disk.
Spinning up the magnetic platters takes extra time and
power. This incurs noticeable delay and power penalty.
Spinning down/up the magnetic platters too often also has a
significant impact on reliability.
29.
30.
31. Disk has to stay in the low-power state for a
sufficiently long period of time which save the
energy . This energy needed to spin the disk up
again.
This techniques is difficult to implement in SSDs
because the energy required to start SSDs is
more than a spanning up the server hard disk.
Decreasing the RPM can significantly reduce the
power consumption of disk drives.
However, lower RPM can further worsen the
performance of large-capacity disk drives.
32. In the past 50 years, disk drive architecture has
remained largely unchanged.
They have reached a turning point at which they
have to be reborn in order to further improve
their performance and reduce power
consumption while still maintaining high
reliability.
Hybrid disk is a temporary approach
◦ Therefore, an architecture shift is required to achieve
this goal.