Transcript SATA in the Enterprise Willis Whittington Sr Mgr Interface Planning Seagate Technology Jeff Mastro Program Manager Microsoft Corporation [email protected] [email protected].

SATA in the Enterprise
Willis Whittington
Sr Mgr Interface Planning
Seagate Technology
Jeff Mastro
Program Manager
Microsoft Corporation
[email protected]
[email protected]
Session Outline
Serial ATA Overview
SATA Goals and Objectives
Interface Enhancements and Features
Compatibility with SAS
SATA Enterprise Applications
Market Segmentation & Tiered Storage
“Nearline”
Caveats for SATA in High End Applications
Error Recovery & Data Integrity
Reliability, Workload, Duty Cycles
Environmental Considerations
Q&A
Session Goals
Provide some insight into the use of high capacity
low cost storage in a tiered storage environment.
Understand the need for a new class of
“Nearline” products which provide storage for
reference data and infrequently accessed data
such as financial compliance archives.
Nearline
High capacity, low cost storage of reference data and data protection
copies that are infrequently accessed, such as financial compliance
archives, medical records and disc-to-disc backup
An appreciation for the challenges of using low
cost storage in an Enterprise computing
environment
Storage Devices: Key Windows Strategies
Storage Fabrics
Server/Enterprise
Personal Storage
Client/Consumer
Optical Platform
Client/Consumer
Preferred
Storage Platform
Partner/Customer
Leading platform supporting storage fabrics
Optimized platform features enabling new
customer scenarios in Personal Storage
Timely, comprehensive, quality platform
support for optical devices
Preferred platform for developing,
integrating, and deploying storage devices
Serial ATA Overview
New topologies, new features, low cost.
Moving from Parallel ATA to Serial ATA
Summary of Goals and Objectives
Primary inside the box storage connection
S/w transparent to ATA
Better cabling (1m) and connectors
and a lower pin count
PATA
(18”)
Higher performance, scalability
and ~10 year roadmap
150  300  600 MB/sec
Support for 1st Party DMA
SATA (1m)
Target can initiate data transfers to host memory
Plug and Play with no jumpers or terminators
Connecting to the I/O
Connectors and cables are required to:
Support 1.5 Gbps with headroom for 3.0 Gbps speed
Be cost competitive to Ultra ATA
Facilitate a smooth transition to Serial ATA
Be suitable for both 3½” and 2½” storage devices
Support up to 1 meter cable length
Be blind-mateable and hot-pluggable (staggered pins)
Data
Rx Tx
3.3V
Voltage Pins
5V R 12V
Key
Gnd
Gnd
Additional SATA II Features
Backplane Interconnect
Staggered spindle spin up
SATA Native Command Queuing
Performance improvement
Non 512 byte block size
Multiple block size support is not
financially viable
Port Multiplier
Provides multi drive access to a host
Port Selector
Provides a failover path to a drive
Port Multiplier
Connects 1 Host to any one of
15 (max) SATA devices
JBOD
Host command selects Target
device
SATA
Avoids changing
cables to change
Targets
Port
Contains switching/processing
Multiplier
On
electronics
backpanel
Requires HBA support
Needed to support multi-drive
cabinets if no SAS Expanders available
SATA Port Selector
Provides an alternate path to the Target in the
event of a failure
Switching between hosts is tantamount to a cable
change at the drive (I/O Initialization required)
Active
Connection
Not to be confused with Dual Port
Port Selector
Passive
Connection
(1 per Drive)
SATA Drive
Compatibility with Serial Attached SCSI (SAS)
SATA is plug compatible with SAS
Connector Flip Side
Port B
SAS
Keyway
SATA
SATA
Pluggable
SAS
☻
Compatibility with Serial Attached SCSI (SAS)
SATA Compatibility with SAS provides distinct
user advantages
SATA drives are plug compatible with SAS backplanes
SATA & SAS drives may be freely intermixed within an
enclosure
Serial Tunneling Protocol (STP) allows a SAS Host to
communicate directly with a SATA Target
SATA devices are addressable in a SAS Expander topology
STP
SATA
SSP
Create Tunnel to Target Drive
Serial
SATA
Serial
from
SCSI
Host
Tunneling
Protocol
to
Target
Protocol
SAS Drives
SATA Drives
SATA Enterprise Applications
Nearline Storage: High Capacity, Low Cost Opportunities
Nearline Disc Drives
“Only a small percentage of data should be maintained on the highest cost,
high performance devices. The rest is migrated to less expensive storage
where it is immediately available when needed.” EMC Whitepaper April 2005
Transactional
Reference Data
Staging &
spooling
Archive
On Line
Near Line
Off Line
Performance
Availability
Access time
High capacity
Low cost-per-gigabyte
High reliability
Optimized for nearline applications
Cost per Terabyte
Capacity
Retention
Workload Descriptions
Transactional Database (ES)
Separately attached storage supporting 1000s of users
Mission Critical Applications e.g. Bank records, ticketing, etc.
General Purpose Server (ES)
Web, Application, E-Mail server for 10-100 users
Small databases and RAID.
Nearline
High capacity, low cost storage
Reference data, Sarbanes---Oxley
Disc-to-disc backup.
Low Cost Server (PS)
Departmental filing, printing, applications
Small web or E-mail server
Desktop (PS)
Average office user
E-Mail, browsing, reading / writing documents, printing.
Typical Workloads by Application
Nearline workloads are unique compared to online and low-cost
server workloads
Online
Highly Transactional Database
General Purpose Server
Nearline
Reference data
Highly random
High IOPS
24x7 power-on
hours
Disc-to-disc backup
Highly random
Low Cost Server
Low IOPS
24x7 power-on
hours
Desktop
High duty cycle
Low-Cost Server
Low duty cycle
8x5 power-on hours
More sequential
Frequent start/stop
Low duty cycle
Data Is Changing – and So Is Storage
Worldwide Reference Information Capacity Forecast
Server attached Disk Storage
Worldwide, Digitized Reference Information created and stored will surpass NonReference Information by the end of 2006
2,500
2,250
2,000
Petabytes
1,750
1,500
1,250
1,000
750
500
250
0
2001
2002
Non-Reference Information
Source: Enterprise Storage Group
2003
2004
Reference Information
2005
2006
Total Storage Capacity
Caveat
The right storage for the right application
Nearline/Enterprise Storage Differences
Cost
Capacity
Duty Cycle
Reliability
Scalability
Environment
Nearline
Storage
7,200 rpm
SATA
Performance
Data Integrity
Power Consumption
System Architecture
Enterprise
Storage
15,000 rpm
SAS/FC
Each application gets the parameters it needs
Nearline / Enterprise Drive Differences
Electronics
Motor
Discs
Head
Misc Mechanical
Stack
Dual
Higher
More
Larger
processors
platters
heads
rpm
Magnets
Multi
Less
Smaller
Low
Stiffer
mass,
host
runout
Covers
diameter
high rigidity
Dual
More
Full
Higher
Air Control
media
port
expensive
costcertification
Devices
design
Twice
Fully
Fastercharacterized
the
Seeks
firmware
High
Low RV
rpmsusceptibility
control & RPS
Command scheduling
Superior error correction
Smart servo algorithms
Perform. optimization
Data integrity checks
ATA Workload vs. Failure Rate
Effect of Duty Cycle on ATA Drive Failures
Cum %
Failures
%Failures
Cumulative
8.0%
Low End Server (Random)
7.0%
6.0%
5.0%
High Duty Cycle (Sequential)
4x
4.0%
3.0%
2x
Desktop Workload
2.0%
1.0%
0.0%
0
100
200
300
400
500
600
700
800
900
1000
1100
1200
Test Time, Hours
HDT
HDT_2
Desktop1
3 groups of 300 Desktop drives were tested in commercial test
chambers
All workloads were captured from exhaustive Field analyses
In repeated testing, greater than 2x higher failure rates were
recorded with Enterprise workload compared to Desktop
UER* on Very High Capacity RAID Sets
The UER for SATA products is 1 in 1014 bits read
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
This means there is a 20% probability of an
Unrecoverable Error during the rebuild.
Better odds would be available with RAID 1 or 6
Higher redundancy, faster Rebuild, more expensive
Risks can be reduced with good error management
Intelligent rebuild (e.g. ignore unused capacity)
Background media scan
*Unrecoverable Error Rate
Nearline UER Comparison to Enterprise
Probability of Unrecoverable Errors during RAID Rebuild
20%
Probability of UE
16%
12%
SATA NL Drives
UER = 10-14
8%
UER = 10-15 in 2007
4%
3
4
5
# of 500 GB Drives in RAID Set
Enterprise Drives
UER = 10-15
UER = 10-16 in 2007
Alternate Sector Size – Not an Easy Proposition
Standard 512 byte Format
Servo
Data 1
Data 2
Data 3
Servo
Data 4
Non-Standard Format (allows for post amble check field)
Servo
Data 1
Data 2
Data 3
Servo Data 3
Data 4
Industry has standardized on 512 byte sectors
H/w and s/w optimized for efficiency, cost and µP loading
Manufacturing geared up for “vanilla” processes and production
Problems with changing the sector size:
More buffering required to handle the additional data
More complex f/w requiring additional memory storage
More complex split sector handling/computation required
More complicated/costly media flaw mapping process
Physical vs Logical Flaw Mapping
Flaw somewhere in LBA 9
Flaw in LBA 7 or LBA 8
Flaw @ Cyl x, Hd y, Byte z
Cyl
Flaw in LBA 7
Index
 Logical flaw location
is format dependent
 LBA can be computed
from the physical loc.
for any media format
 Physical media certification
is a complicated operation
 Complex algorithm used
 Time consuming Manuf. Process
 Costly
Reformat
to 528
512
byte sectors
byte sectors
End to End Data Checking
520 Byte Sector
Data Frame
SOF
Header
User
Logical Block Logical Block
Application
Guard
Tag
2 Bytes
8 Bytes
512 Bytes
2 Bytes
End to End
CRC EOF
Data Check
Data
Logical Block
Reference
Tag
4 Bytes
Not available on SATA drives with fixed 512 byte sectors
Enterprise Drive Data Integrity Provisions
I/O
Interface Electronics
Disk Formatter
Customer Data (512)
+
CRC
Customer Data
+
Err. Det.Check
+
PBA Check
512 bytes data
+
40 check bytes
Data Buffers
Customer Data
+
Err. Det.Check
+
LBA Check
LBA = Logical Block Address
PBA = Physical Block Address
Protection against hardware failures in the
data path
Effectively eliminates the risk of unreported
“data miscompare” errors
Rotational Vibration
Queue = 4
200
Mutual Interference
180
Increases seek time
160
Takes longer to settle on track
140
Actuator vibrates off track
Enterprise
120
Performance (IO/sec)
IO/s
Read retries
Aborted Writes
Enterprise Requirement
100
80
60
40
20
Desktop
0
0
20
40
60
-20
Rotational Radians/sec**2
Vibration (Rads/sec2)
80
Rotational Vibration
Average RV recorded on 33 Enterprise Cabinets
Very
Bad
40
R V (Rads/sec2)
35
30
25
20
Enterprise
15
10
5
Very
Good
0
Desktop
Cabinets Tested
More stringent RV spec. needed for SATA cabinets
RV aggravated by random operation and “bursty” workloads
Summing Up ….
SATA enables a Tiered Storage approach to data
management by providing a high capacity low
cost solution for what has come to be called the
Nearline segment of the market.
In the hierarchy of storage, Nearline provides
continuity by filling, and overlapping, the slot
between high end Personal Storage and
Enterprise
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.
and in Conclusion…
Call To Action
Use RAID 1 or RAID 6 in SATA Array
Uncorrected Error Rate Mitigation
Extra cost justified by value derived
Rotational Vibration Consideration
Exaggerated by Random Disk Operations
Deploy SATA drives in a high-quality cabinet
Extra cost justified by value derived
Remember the Duty Cycle
Optimal Nearline Drive Application
Sequential Read Operations (WORM)
Compliance, Email, Customer Record Archive
Don’t Expect Desktop Drive to Handle Enterprise Load
Additional Resources
Web Resources:
Specs: http://www.t13.org/#Standards
Whitepapers:
http://www.microsoft.com/whdc/device/storage/serialATA_FAQ.mspx
http://www.sata-io.org/docs/serialata%20%20a%20comparison%20with%20ultra%20ata%20technology.pdf
http://www.lsil.com/files/docs/marketing_docs/storage_stand_prod/Te
chnology/SATA_II_White_Paper.pdf
Questions on this topic
[email protected]
General WinHEC 2005 Comments
[email protected]