SSDs Driving Greater IT Efficiency
June 2011
Stephen Weinger – Director of Marketing
Vince Kim – Sr. Technical Manager
2011.04.13/NAND MKT(Memory)
Beyond Mega and Giga…
What is the highest defined unit of measurement
(Metric Prefix)?
2011.04.13/NAND MKT(Memory)
Beyond Mega and Giga…
What is the highest defined unit of measurement
(Metric Prefix)?
SI Prefix Symbol
10n
Short Scale Decimal
None
None
1
One
1
Deca-
Da
101
Ten
10
Hecto-
H
102
Hundred
100
Kilo-
K
103
Thousand
1,000
Mega-
M
106
Million
1,000,000
Giga-
G
109
Billion
1,000,000,000
Tera-
T
1012
Trillion
1,000,000,000,000
Peta-
P
1015
Quadrillion
1,000,000,000,000,000
Exa-
E
1018
Quintillion
1,000,000,000,000,000,000
Zetta-
Z
1021
Sextillion
1,000,000,000,000,000,000,000
Yotta-
Y
1024
Septillion
1,000,000,000,000,000,000,000,000
?
2011.04.13/NAND MKT(Memory)
Beyond Mega and Giga…
What is the highest defined unit of measurement
(Metric Prefix)?
SI Prefix Symbol
10n
Short Scale Decimal
None
None
1
One
1
Deca-
Da
101
Ten
10
Hecto-
H
102
Hundred
100
Kilo-
K
103
Thousand
1,000
Mega-
M
106
Million
1,000,000
Giga-
G
109
Billion
1,000,000,000
Tera-
T
1012
Trillion
1,000,000,000,000
Peta-
P
1015
Quadrillion
1,000,000,000,000,000
Exa-
E
1018
Quintillion
1,000,000,000,000,000,000
Zetta-
Z
1021
Sextillion
1,000,000,000,000,000,000,000
Yotta-
Y
1024
Septillion
1,000,000,000,000,000,000,000,000
?
2011.04.13/NAND MKT(Memory)
Beyond Mega and Giga…
What is the highest defined unit of measurement
(Metric Prefix)?
SI Prefix Symbol
10n
Short Scale Decimal
None
None
1
One
1
Deca-
Da
101
Ten
10
Hecto-
H
102
Hundred
100
Kilo-
K
103
Thousand
1,000
Mega-
M
106
Million
1,000,000
Giga-
G
109
Billion
1,000,000,000
Tera-
T
1012
Trillion
1,000,000,000,000
Peta-
P
1015
Quadrillion
1,000,000,000,000,000
Exa-
E
1018
Quintillion
1,000,000,000,000,000,000
Zetta-
Z
1021
Sextillion
1,000,000,000,000,000,000,000
Yotta-
Y
1024
Septillion
1,000,000,000,000,000,000,000,000
And
Finally?
2011.04.13/NAND MKT(Memory)
Beyond Mega and Giga…
What is the highest defined unit of measurement
(Metric Prefix)?
SI Prefix Symbol
10n
Short Scale Decimal
None
None
1
One
1
Deca-
Da
101
Ten
10
Hecto-
H
102
Hundred
100
Kilo-
K
103
Thousand
1,000
Mega-
M
106
Million
1,000,000
G of all10
Billion
TheGigacombined space
hard drives
in the world1,000,000,000
does not amount to one Yotta-byte
9
TeraPeta-
T
1012
Trillion
1,000,000,000,000
What
comes
next???
P
10
Quadrillion
1,000,000,000,000,000
15
Exa-
E
1018
Quintillion
1,000,000,000,000,000,000
Zetta-
Z
1021
Sextillion
1,000,000,000,000,000,000,000
Yotta-
Y
1024
Septillion
1,000,000,000,000,000,000,000,000
2011.04.13/NAND MKT(Memory)
What Comes Next? (10 ^ 27)
Goog-a-Byte?
Zuck-a-Byte?
Sheen-a-Byte?
Hella-Byte?
Actual Petition for Hella as 10 ^ 27
> 63,000 Facebook Fans
International Committee in France to Decide
Chances Appear Hella Slim…
2011.04.13/NAND MKT(Memory)
Agenda
1. IT Trends
2. SSDs Value in Enterprise Systems
3. SSD Technology Trends
2011.04.13/NAND MKT(Memory)
Mobile Internet will be 10x Desktop Internet
Units, M
CE Devices:
Computing Growth Drivers Over Time
Beyond Phones
Mobile
Internet
100,000
10,000
Wired
Internet
1,000
1B+
Units
100
Minicomputer
10M+
Units
10
Mainframe
Increasing Integration
1
1M+ Units
PC
100M+
Units
>10B
Units
Stand-Alone
Wired
Seamless/Cloud
New Eco-Systems
1960
Source: Morgan Stanley
1970
1980
Back-end Infrastructure:
Cloud Computing
1990
2000
2010
2020
2010
2011.04.13/NAND MKT(Memory)
Impact of Facebook Keeps Growing…
Steve
Weinger
Internet Rank
2009
#3
2010 Y/Y
#2
+1
WW Active Users 250M >500M 2X
Use Daily
120M
250M
2X
Photos / Month
1B
>3B
3X
Videos / Month
10M
>20M
2X
Mobile Access
30M
>200M 7X
People that use Facebook on their mobile devices are twice as
active on Facebook than non-mobile users – Facebook.
2011.04.13/NAND MKT(Memory)
Web 3.0 Impact on Traffic Load
Ever-Increasing Data Traffic
Load at 40% CAGR
Profound Impact on
Infrastructure Requirements
55.6K
Petabytes
Per Month
42.2K
30.8K
21.4K
WEB
3.0
14.7K
10.2K
2008
2009
2010
2011
2012
2013
The Cisco Visual Networking Index
1PB = 13.3 Years of HD Video or 4.7B eBooks
2011.04.13/NAND MKT(Memory)
Data Center Power Consumption Growing
Data Centers Account for 23% of Global ICT Power Consumption
(About 1% of Total WW Power – and Growing…)
Data Center
(23%)
440TWh
822TWh
220TWh
Global ICT
Power Consumption
108TWh
ICT accounts for 3.7% of the
Total Global Power Usage
Data Centers
*
Netherlands
Australia
South Korea
(Information Communication Technology)
2011.04.13/NAND MKT(Memory)
Challenge: Data Center Carbon Footprint
CO2 Emissions as % of WW total
WW Data Center Emissions,
(metric megatons CO2)
CAGR >11%
0.6
0.8
1.0
340
0.2
Data
Centers
Airlines Shipyards
Steel
Plants
>4X
80
2007
Source: McKinsey, 2008
2020
2011.04.13/NAND MKT(Memory)
Challenge: Data Center Power Consumption
1W Power Requires 2.5W Cooling
2009 Server Expenditures: $43B, another $33B in Power
> 70% of Power Bill in Financial Services Market
Source: EPA ‘07, IDC, Green Media
2011.04.13/NAND MKT(Memory)
SSD Market Forecast
 SSD is expected to show strong growth
–
Reaching 78M units and $8.5B by 2014
SSD Total Shipment
SSD Total Revenue
78M
$8.5B
Server & Storage
Server & Storage
PC
PC
$5.6B
40M
$3.7B
$2.5B
23M
$1.6B
14M
8.0M
2010
2011
2012
2013
2014
2010
2011
2012
2013
2014
Source: Gartner Q4 ‘10
2011.04.13/NAND MKT(Memory)
Enterprise SSD Market Forecast
 Enterprise SSD Market is still in Emerging Status
– Enterprise SSD shipments are <10% of total SSD market but
revenue is >30% of total SSD market
SSD Total Shipment
SSD Total Revenue
$3.5B
7.2M
Storage
Storage
Server
Server
$2.3B
3.9M
$1.6B
$1.1B
2.1M
0.6M
2010
$0.7B
1.1M
2011
2012
2013
2014
2010
2011
2012
2013
2014
Source: Gartner Q4 ‘10
2011.04.13/NAND MKT(Memory)
SSDs Value in Enterprise Applications
2011.04.13/NAND MKT(Memory)
What is SSD?
SSD is Solid State Drive
 Solid State Drive is a large capacity of Storage using NAND
Flash Memory as its media
 For interface wise, SSD is
using S-ATA/SAS/FC
interface for compatibility to
conventional industry and
PCIe interface for lower
latency
2011.04.13/NAND MKT(Memory)
Hardware: SSD vs. HDD
 SSD doesn’t have
SDRAM
– Mechanical component
• Head, platter, motor
– Read- channel, Preamp
 SSD does need
– FTL (Flash Transition Layer)
– More PCB Layers for better
SI/PI control
Media
PreAmp
SOC
NAND
NAND
NAND
NAND
NAND
NAND
NAND
SOC
NAND
Read-Channel
Hard Disk
Controller
Spindle
motor
VCM
SATA
COMBO
2011.04.13/NAND MKT(Memory)
SSD Values: Reliability
 SSDs offer higher reliability vs HDD
– Lower maintenance costs
Environmental Spec
NAND flash
Controller
Connector
HDD vs. SSD
MTBF
1.5Mhrs vs 2Mhrs
Shock
60G vs 1500G
Vibration
1.2G vs 20G
Acoustics
3.1bels vs 0bels
Source: Available datasheet
*G : gravity
*bel : sound power
2011.04.13/NAND MKT(Memory)
HDD Latency is creating performance bottleneck
 15K HDD IOPS is too low due to high latency
 The Performance Gap is Widening  How do you fix?
2011.04.13/NAND MKT(Memory)
Fixing the Performance Bottleneck
 HDDs cannot spin faster  Stagnant IOPS Growth
 SSD performance improves with each new generation
4 core
6 core
10 core
CPU
DDR2
10Krpm
SAS HDD
DDR3
15Krpm
SAS HDD
DDR4
Memory
SSD
DISK
2011.04.13/NAND MKT(Memory)
SSD Values: Power Dissipation
 SSD uses much lower power than 15K SAS HDD
– In case of active mode, 55% power saving
– In case of idle mode, 76% power saving
15Krpm SAS HDD
SSD
15Krpm SAS HDD
SSD
[Source : Storage review.com]
2011.04.13/NAND MKT(Memory)
SSD Values: High Performance
 SSD shows 110X better IOPS than 15K SAS HDD
15Krpm SAS HDD
SSD
[Source : Storage review.com]
2011.04.13/NAND MKT(Memory)
SSD Values: Power Efficiency - IOPS/Watt
 SSD widens the gap in power efficiency
15Krpm SAS HDD
SSD
[Source : Storage review.com]
2011.04.13/NAND MKT(Memory)
SSD Values in System: TPC-C
 SSD Performance improves as the number of users increases
– 6X improvement on TPC-C Benchmark
2011.04.13/NAND MKT(Memory)
SSD Values in System: TPC-C
 SSDs allow for 6X more transactions per minute.
Translating TPC-C result to Real World
11K customers
w/ 2 ea SSD
per minute
Stock level
Delivery
Order Status
Payment
New Order
Just replacing HDD with SSD, take 6X more customers.
1.8K customers
w/ 8ea 15Krpm SAS HDD
per minute
Stock level
Delivery
Order Status
Payment
New Order
[Source : Samsung Lab]
2011.04.13/NAND MKT(Memory)
SSD Values in System: Storage
 SSD increases system performance while decreasing power & cost
IOPS Performance
38%
UP
Power consumption[kW/yr]
Storage Cost[$]
32%
Down
99,300
17%
Down
30,960
10.23
25,690
72,000
6.96
HDD
System
SSD
System
HDD
System
SSD
System
HDD
System
SSD
System
* HDD System: 240 x 300GB 15K FC Disks
* SSD System: 8 x 200GB SSD + 120 x 400GB 10K FC Disks + 24 x 1TB SATA
* Source: Samsung and customer data
2011.04.13/NAND MKT(Memory)
SSD Values in System: Web Service
 SSD based server can enhance work efficiency up to 60%
• Source : Standard Performance Evaluation Corporation [SPEC.org]
•* Assume : ecommerce and e-banking server are accessed for
24hrs
2011.04.13/NAND MKT(Memory)
SSD Values in System: Virtual Machine
 SSD based server improves job completion time up to 90%
[Virtual Machine Demo@
[Data de-duplication]
]
[Data Indexing]
290GB Database Processing Time
[Unit : minutes]
70mins
90%
Time Save
7mins
4 MS SQL Server on virtual machine
15K RPM
HDD based server
SSD based server
[Test Condition]
Model : HP DL385 G7
Storage Option : 1) 120GB Samsung SSD 2) 146GB 15K rpm Enterprise HDD
Test Process : 1) Executing 4 of virtual machine on one HP DL385 G7 2) Executing SQL workload for 290Gb database on each virtual machine
3) Comparing job completion time between SSD based server and HDD based server
2011.04.13/NAND MKT(Memory)
SSD Values in System: Virtual Machine
IOPS
4 VM Comparison SSD vs HDD
SQLIO 8KB Random Reads
Que Depth
Source – HP Aug’10
2011.04.13/NAND MKT(Memory)
SSD Values in System: Virtual Machine
 SSDs remove the IO Bottleneck
IOPS
4 VM Comparison SSD vs HDD
SQLIO 8KB Random Reads
Que Depth
Source – HP Aug’10
2011.04.13/NAND MKT(Memory)
SSD TCO: “Instantaneous Break-even”
Source: Storagereview.com, March, 2009, IOPS calculated by IOMeter File Server average 1 I/O to 128 I/O w/ RR70% and RW30%
Source : HP website / www.hp.com/go/thermallogic)
2011.04.13/NAND MKT(Memory)
SSD Technology Trends
2011.04.13/NAND MKT(Memory)
Tech Trend I: Sustained Performance
 Push for sustained performance
– Sustained performance means supporting stable service without
fluctuation
– FW optimization, over- provisioning and battery back- up
technology keep improving end user experiences by enabling
sustained performance
– Eco- system (OS, File system) is being optimized for SSD
Before
After
FW
optimize
2011.04.13/NAND MKT(Memory)
Tech Trend II: TBW (Terabyte Written)
 TBW is the number of total TeraBytes Written to the SSD
– Mapping algorithm, wear leveling scheme, write amplification
factor and the program/erase cycling capability of the NAND
components impacts TBW
 That is, TBW means the lifetime of SSD
The Concept of TBW
Amount of Nominal
Data Write
Amount of Physical
Data Write
TBW
• Mapping Algorithm
• P/E Cycle capability of
• Wear Leveling
the NAND Component
• Write Amplification Factor
• SSD capacity
2011.04.13/NAND MKT(Memory)
Tech Trend III: Performance Throttling
 ‘Performance Throttling’ technology could be utilized for
differing lifetime expectations from various customers
– Static performance throttling
• Suggest lower but more sustained performance to meet fixed lifetime
– Dynamic performance throttling
• Suggest variable performance according to customer’s request
Static Performance Throttling
Dynamic Performance Throttling
Target Performance for 5yrs
Static Performance
0
1
2
Dynamic performance in real usage
Performance
Performance
Peak Performance
3
Usage[Year]
4
5
0
1
2
3
4
5
Usage[Year]
2011.04.13/NAND MKT(Memory)
Tech Trend IV: High Performance NAND
 High Performance NAND
– Step 1 : Performance improvement with toggle mode NAND
• NAND Interface Speed: 66Mbps  133Mbps  400Mbps
Random Read (IOPS)
Random Write (IOPS)
20%
10%
40%
66Mbps SDR
133Mbps DDR
400Mbps DDR
66Mbps SDR
60%
133Mbps DDR
400Mbps DDR
– Step 2 : Lower power consumption with low power NAND
• IO operating voltage (Vcco): 3.3V  1.8V
– Step 3 : Enterprise MLC NAND into enterprise grade SSD with
extended P/E cycles
2011.04.13/NAND MKT(Memory)
2011.04.13/NAND MKT(Memory)