Desktop, Nearline &
Enterprise Disk Drives
What’s the difference?
Willis Whittington, Seagate Technology
Feb 21st, 2008
Abstract
Desktop, Nearline & Enterprise Disk Drives
What’s the difference?
For the past twenty five years the storage marketplace has been
divided into two major categories namely “Desktop” and
“Enterprise”. Recently, a third player variously known as “Nearline”,
“Reference” or “Business Critical” has evolved to provide a low cost,
high capacity storage solution for Enterprise data that no longer
needs to exist in a high availability transactional processing
environment but must maintain 24 x 7 availability as a reference or
backup resource.
Each of these classes of drives requires a unique and specific set of
attributes to fulfill its role. This presentation will explore these
differences and explain why you need to use
the right drive for the right application.
Agenda
Basic Comparisons
SAS & SATA Compatibility
The Advantages of Nearline SAS
Rotational Vibration
Data Error Rate
Error Correction Capability
Data Integrity
Performance
Annualized Failure Rate
~ Q & A along the way ~
Basic Comparisons
Comparison Table DT / NL / MC*
Key:
Metric
Good
Better
Best
Desktop (DT)
SATA Nearline (NL)
SAS Nearline (NL)
Enterprise MC*
Capacity (GB)
1,000
1,000
1,000
450
Cost
Low
Mid
Mid
High
1x
1.2x
1.2x+
1.5x
600,000
1,200,000
1,200,000
1,600,000
Low (<10%)
Low/Medium (<20%)
Low/Medium (<20%)
High (100%)
Parity (?)
EDC + (ECC?)
EDC/ECC + Proprietary
Data Integrity Protection
EDC/ECC + Proprietary
Data Integrity Protection
Unrec Error Rate
10-14
10-15
10-15
10-16
RV Radians/sec2
6
12.5
12.5
>21
SATA
SATA + Time Control
Full SCSI
Full SCSI
Firmware/Features
Standard
SATA
SATA + Selected
Nearline Features
SCSI + Adv. Features
SCSI + Adv. Features
(Enabled by Dual CPU)
(Enabled by Dual CPU)
Power On Hrs/Year
2400
Power Consumption
MTBF (Hrs)
Duty Cycle
Data Integrity
Error Recovery
8,760
8,760
(Low Duty cycle)
(Low Duty cycle)
8,760
Multi Initiator
No
No
16 Hosts & Dual Port
16 Hosts & Dual Port
Performance
1x
1x
1x+
1.4x / 2.5x (Seq / Rand)
T10 Data Protection
No
No
Yes
Low
Low
High + Dual Port
Scalability
*Mission Critical
2x Duplex
Yes
High + Dual Port
2x Duplex
Anatomy of an Enterprise Drive
Motor
Higher rpm than NL or DT
Tighter specifications
Less runout
More expensive
Anatomy of an Enterprise Drive
Discs
Four platter design
Smaller diameter than NL/DT
Full media certification
Fully characterized
Variable sector format
Anatomy of an Enterprise Drive
Head Stack
Eight head design
Low mass, high rigidity
Voice coil designed for
o optimal performance
o 100% duty cycle
Higher cost design
Anatomy of an Enterprise Drive
Environmental
Control Module
Humidity Control
Chemical Absorbtion
Multi-point filtration
Windage Design
Anatomy of an Enterprise Drive
Misc Mechanical
Powerful Voice Coil Motor
Stiffer Covers
Air Control Devices
Faster Seeks
High Servo Sample Rate
Low RV susceptibility
Anatomy of an Enterprise Drive
Electronics
Dual processors
Multi host queuing
Dual port
Twice the memory of NL/DT
High rpm control
Command scheduling
Superior error protection
Superior error correction
Smart servo algorithms
Perform. optimization
Data integrity checks
Sequential h/w assist
Con>>
SAS & SATA Compatibility
I/O Connectors for SAS & SATA
Vcc
Enclosure
Vcc
Drive
Enclosure
Vcc
Drive
Vcc
~3msec
Current Limited
Steady State
Data
Rx Tx
3.3V
Voltage Pins
5V
12V
LED/
Spin-up
Key
Gnd
For SAS, the key-way is
filled in and its flip side
is used for the 2nd Port
<< Con
Gnd
This prevents a SAS drive from being
plugged into a SATA cabinet slot.
The Advantages of
Nearline SAS
NL SATA Compared to NL SAS
Stepping up to SAS
provides
Mission Critical Compatibility
NEARLINE SAS
Dual Port
Full EDC & ECC
100% Phy Compatible
End-to-End Data Integrity
Variable Sector Size
Multiple Host Support
Full SCSI Command Set
Enterprise Command Queuing
Concurrent Data Channels
Full Duplex (Bidirectional) I/O
NEARLINE SATA
SAS/SATA NL Physical Differences
SAS
Electronics
SAS
Port “B”
SATA
Electronics
Nearline
Head/Disc Assy.
SAS/SATA NL Differences
Dual Port
Two Data Channels
2 Concurrent Writes
OR
2 Concurrent Reads
OR
1 Write + 1 Read
SAS
CHANNEL 2
Full Duplex
CHANNEL 1
SAS
SATA
Xmits
In both
directions
at one time
Half Duplex
One Data Channel
1 Write
OR
1 Read
SATA
Xmits
In one
direction
at one time
Multi Host Command Queuing
SAS
SAS
EXPANDER
EXPANDER
SAS
Drive Queue
Supports
16 Hosts
Interposer
Interposer
to handle Q’ing
for 2nd host
SATA
Drive NCQ*
supports a
single Host
*Native Command Queuing
NL SAS/SATA Summary
Both SATA and SAS Nearline drives are designed for
use in Enterprise Mission Critical environments.
SAS Nearline drives have additional advantages which
are made possible by the Serial SCSI interface and
enterprise electronics:
• Full system interface compatibility at the protocol, physical
(“phy”), and command level
• Enterprise error recovery and performance optimization
controls
• Full data integrity protection both within the drive and at
the system level with DIF support.
Rotational Vibration
RV Emitted by a Seeking Drive
• PS drives are not designed for
Neighboring Drive’s
backplane (JBOD/SBOD) use
Servo needs to
and are not equipped to cope
compensate for
with the effects of RV
externally
induced RV
RV is Proportional to Seek Current
Current in Voice Coil
Disc Movement
Coil Current
‘Scope Picture, Seagate Prod. Dev.
HDA subjected to rotational forces
Rotational Vibration
Impact on Performance*
6
12.5
21 rads/sec2
-
120
Enterprise
Enterprise
80
Barracuda ES
Enterprise
40
Nearline
60
Desktop
% Degradation
100
Nearline
Desktop
20
Desktop
0
0
10
20
30
40
RV Level (Rad/sec2)
50
60
*Source: STX Competitive Analysis.
RV in 33 Different Cabinets
45
Rotational Vibration
40
2
rads/sec (rms)
11 Unacceptable
35
Acceptable for Enterprise Nearline
30
22
Desktop
12
5
25
Enterprise
21 Rad/sec2
20
15
SATA Nearline
12.5 Rad/sec2
10
SATA Desktop
6 Rad/sec2
5
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Cabinet Tested
20
21
22
23
24
25
26
27
28
29
30
31
32
33
More stringent RV spec. needed for SATA cabinets
RV aggravated by system fans, random access and “bursty” workloads
Data Error Rate
UER* on High Capacity RAID Sets
The UER for SATA desktop is 1 in 1014 bits transferred
• 1014 bits = 12½ terabytes
A 500 Gbyte drive has 1/25 x 1014 bits
Rebuilding a SATA drive in a RAID 5 set of 5 drives means transferring
5/25 x 1014 bits = 1/5 of UER spec.
• 20% probability of an Unrecoverable Error during the rebuild.
Better odds would be available with RAID 1 or 6
• RAID 1 rebuilds from a single mirror drive
• RAID 6 can tolerate a second error during the rebuild.
Risks can be reduced with good error management
• Intelligent rebuild (ignore unused capacity)
• Background media scan (dynamic certification)
*Unrecoverable Error Rate
DT / NL / MC UER*
20%
Probability of Unrecoverable Errors during RAID Rebuild
Probability of UE
16%
Desktop Drives UER = 10-14
12%
8%
4%
Nearline UER = 10-15
2%
0.2%
3
4
5
# of 500 GB Drives in RAID Set
*Unrecoverable Error Rate
Enterprise UER = 10-16
Error Correction Capability
Standard vs Reverse ECC
(Write Command)
+
User ec
randomized
ec
User
User Data
ECC Generator
User
User Data
Randomizer
RLL Encoder
+
randomized
RLL Encoder
Randomizer
encoded
encoded ec
ec
encoded
ECC Generator
Standard vs Reverse ECC
(Read Command)
_
ec
??
ECC Correction
De-randomize
ec
120 Bit Error
(Propagates
in Decoder)
20 Bit Error
RLL Decoder
User Data
_
De-randomize
User
User Data
20 Bit Error
ec
No Error Propagation
RLL Decoder
ec
ECC Correction
Error Corrected on the fly
Sync Mark Errors on SATA Drives
Sync Mark
Sync Field
Media Flaw
Sector Format
User Data Field
ECC
ƒ The Sync Field is used to get the read channel in
frequency sync with the data recorded on the media
ƒ The Sync Mark is used to define the beginning of the User
Data Field
ƒ Failure to recognize the Sync Mark (due to a thermal
asperity or a grown media defect) means the User Data
Field is not delineated and the data is lost.
Sync Mark Errors on SAS Drives
Sync Mark
Sync Field
Media Flaw
n bytes
Sync Mark 2 embedded in the data field
User Data Field
Sector Format
ECC
Read Operation
????? !!!
Read Channel realizes the
first Sync Mark is missing:
- Loads Buffer with n zeroes
- Starts searching for SM2
0011011000011110101011000011010100110110001010101000111 ECC
Read Channel finds SM2
and reads the data following
it into the Buffer, starting at
location n+1
DATA BUFFER
Before ECC
00000000000000 0011011000011110101011000011010100110110001010101000111
The missing n bytes are recovered using the ECC
After ECC
11010111010011 0011011000011110101011000011010100110110001010101000111
Data Integrity
Single Sector Data Transfer from the Host
SOF
User Data
Header
DIF
CRC
EOF
FC or SAS
Interface
Interface
Disparity Check
Odd Parity Generator
=
CRC Generator
NO
=
NO
ERROR
Drive I/O
Checks
DIF Generator
Frame Buffer
CRC
DIF
DIF
Parity
User Data path
=
NO
User Data
Odd Parity Generator
Encryption Engine
IOEDC/IOECC Engine
IOEDC Generator
ERROR
NO
Cache Buffer
User Data
DIF IOEDC
EDC/ECC
IOEDC/IOECC Engine
RLL Encoder
User Data
DIF
IOEDC
DIF
IOEDC ECC
Reverse ECC Generator
User Data
=
Randomizer
YES
User Data
DIF IOEDC
Performance
Performance Comparison
100
Enterprise Sequential Access
90
80
Desktop / NL Sequential Access
Mbytes/sec
70
Vendor Range
60
50
Sequential Transfer Rate α RPM x Disc Dia x Bit Density
40
Enterprise Transfer Rate
Desktop Transfer Rate
30
=
=
20
RPM x Disc Dia x Bit Density
RPM x Disc Dia x Bit Density
15000 x 65
7200 x 95
= 142%
(independent of seek time & Latency)
10
1
4
Q Depth
16
32
Performance Comparison
100
400
90
360
80
320
o
Rand
e
s
i
r
rp
Ente
Desktop / NL Sequential
70
m
280
60
240
50
200
40
160
do
Desktop / NL Ran
30
m
120
20
80
10
40
1
4
16
32
Operations/sec
Mbytes/sec
Enterprise Sequential
SPC-1C Performance Comparison
SPC-1C Benchmark
30
Response
ResponseTime
time (ms) (ms)
25
Desktop
20
Vendor & Model
Variance
Enterprise
15
Desktop/NL
10
Enterprise
5
OLTP
0
0
100
200
IO/s
IO/s
300
400
• SPC-1C comprises I/O operations demonstrating small storage
subsystem performance (1-16 drives) while performing the typical
functions of a business critical application.
500
Database
Email
SPC-1C Workload
AFR
(Annualized Failure Rate)
AFR vs Duty Cycle
Typical Enterprise application, Power-On-Hours, and Temperature
1.2
Desktop Drives
AFR (%)
1
Equivalent Enterprise Duty Cycles
0.8
2½ x AFR
Enterprise Drives
0.6
0.4
0.2
SATA drives in Enterprise applications run hotter, at higher duty cycle,
and for more Power-On-Hours than in desktop applications.
0
0.5
Low
1
2
3
5
7
Duty Cycle (%)
10
15
20
High
In Conclusion…..
Although technological advances, driven
by Enterprise research, will be leveraged
into SATA products, there will continue to
be functional limitations imposed on
these devices by the overriding metric of
Low $/GB Storage.
Download PDF
Similar pages