understanding performance of ssds in the enterprise...understanding performance of ssds in the...

Understanding Performance of SSDs in the EnterpriseWill AkinPrincipal Engineer, Intel NAND Solutions GroupDr. John BuschCo-founder, CEO, Schooner Information Technology

MEMS002

SF 2009

2

Agenda Enterprise Performance Tuning

• Why Intel SSDs in enterprise

• Performance nuances of an SSD

• Schooner –Demonstrating the value of Performance

3


• Why Intel SSDs in enterprise

• Performance nuances of an SSD


DRAM Add lots of

DRAM Store the

working set in DRAM to avoid disk latency

Traditional Storage Optimizations

The Bottom Line

using traditional high performance hard disk drives

HDD Lots of

Spindles Reduce Seeks,

Increase IOPS

IOPS Cost & TCOPower & Energy Costs

Add DRAM & Spindles to address bottlenecks in the server storage hierarchy

4

What if we fill one rack with SSDs?

36,000 IOPS 12 GB/sec

sustained BW 1452 Watts

Per HDD*:– R/W 100 MB/sec– 300 IOPS– 12.1 W (active)

4,200,000 IOPS 36 GB/sec

sustained BW 288 Watts

Per SSD**:– Read 250 MB/sec – Write 170 MB/sec – 35,000 IOPS (Read)– 2.4 W (active)

SSD changing the economics of data centrehigher performance with lower energy cost

120 HDDs 120 SSDs

Read IOPS115X

Increase

Sustained BW 3X

Increase

Energy Costs 5X

Reduction

5

* Typical 143GB 15K RPM HDD**Intel® X25-E SSD 160 GB

6


• Why Intel SSD in enterprise

• Nuances of SSD Performance


Nuances of SSD Performance

Four Elements to SSD Performance

1. Type of NAND• Single Level Cell (SLC)• Multi Level Cell (MLC)

2. Indirection system• Erasing and Writing Blocks

3. Host traffic pattern• Workload and Fullness of SSD

4. Spare area• SSD Workspace

7

8

Nuance of the “Indirection System”

• Logical to physical LBA mapping removes need for atomic operations like read modify write (RMW)– The placement of new LBA information can be packed into pages

that are at new physical locations

• Data placement in previously erased blocks makes foreground work (Host IO operations) faster

• Indirection “clean up” needs to reclaim invalid physical locations in background

HostLogical

LBA

SSDL0 to P0

L0 to P348

L0 to P5120

SSD converts a Physical Page to Logical LBA. Logical LBA will not reside in the same physical location each time it is written

9

Host Traffic Pattern: Empty vs Full• An empty SSDs achieves its

maximum write performance under all workloads

• Once initially filled performance will decrease

• Steady State write performance is achieved when the SSD has settled into a consistent write latencies pattern

• A Steady State can be observed when– User capacity is full– Consistent work load is provided

Sequential

data

Spare AreaSpare Area

SSDs steady state performance will have dependencies on

the amount of spare area

10

Host Traffic Pattern: Sequential vsRandom

• Steady state performance of an SSD full of sequential data is better than the steady state of an SSD full of random data– Sequential sectors will be invalidated in

larger linear clusters than random.– Invalidation of sectors within a block is

spotty in random writes.

• Changing the workload of an SSD from sequential to random will cause the performance to fall whereas changing from random to sequential will increase performance over time.

Seq

uen

tial

data

Ran

dom

data

Spare AreaSpare Area

Valid Data

Invalid Data

Drive Performance vs Spare Area

• As spare area increases so does performance

• MLC has a greater % performance increase due to the relative smaller spare area to start with

Intel Confidential

IOPS scales with increase in spare area

0%

100%

200%

300%

400%

500%

600%

0

2000

4000

6000

8000

10000

7% 17% 27% 47% 57%

IOP

S

% Spare Area

160GB MLC Performance scaling w/ over provisioning

IOPS

%IOPS

0%

10%

20%

30%

40%

50%

60%

70%

0

2000

4000

6000

8000

10000

27% 39% 51%

IOP

S

% Spare Area

64GB SLC Performance scaling w/ over provisioning

IOPS(avg)

% IOPS

13

IOPImprovement

IOP Improvement

14

Summary of Performance Impacts

• NAND type SLC VS MLC

• Indirection system optimization

• Host traffic pattern – empty vs full, sequential vs random

• Spare Area contributes to the write amplification factor

• Efficient Firmware optimizes the use of the NAND, Spare Area, and Workload

NAND SpeedSLC vs MLC

+

Workloadrandom vs sequential Read/Writes

+

Spare AreaSpare Capacity work space

+

Firmware EfficienciesEfficient NAND writes and

wearleveling

Workload and Spare Area are outside influences on performance

15


• Performance Nuances of an SSD

• Impacts on SSD Performance

• Schooner –Demonstrating the Value of Performance

16

Agenda Schooner Overview

Caching Tier

Database Tier

17

Agenda Schooner Overview

18

U.S. data-centers use more energy than the entire nation of Sweden.

- EE Times

Datacenter equipment is only utilized 6% to 10%.

- William Forrest Forbes

The number of installed servers in the U.S. will increase from 2.2 million in 2007 to 6.8 million in 2010.

- Frost & Sullivan

From 2003 to 2008 the data size of the average web page has more than tripled.

–websiteoptimization.com

For every 100 units of energy piped into a data center, only three are used for actual computing.

- U.S. Department of Energy

Too Much Rack, Power, and Pipe

19

WebTier

Web App Servers

WWW

Typical Web 2.0 and Cloud Deployment

End User

CachingTier

! Memory Bound

StorageTier

Po

ol 1

Cache Servers

Po

ol 2

DatabaseTier

MasterSlaves

! Disk Bound

20

Schooner Balanced Platform Hardware

• Hardware Features– Dual quad-core Intel® Xeon®

processors 5500 series

– ½ terabyte of Intel® X25-E Extreme SATA Solid-State Drive flash memory

– 64 GB DRAM

– 1/10 Gb Ethernet

– Built from the ground up for reliability and serviceability; all critical components are field-replaceable

– IBM X-Series M2 3650 server platform

64GBDRAM

Dual Xeon5500 Processors

PCleHUB

Parallel FlashControllers

10GbE(Memcached appliance only)

8x MultiGbE

ExternalNetworking

SSDs

21

AdminConfigure Monitor Control

Optimize

Flash Access:Asynchronous I/O Handling Data Striping Interrupt Batching

Flash Management Subsystem:Space Allocation and Shard Management

Object Replacement (cache mode) Persistency Management Tiered Storage Management

Data Fabric:Object Attributes Thread and Core Management

Synchronization/Concurrency Management DRAM Cache ManagementContainer Management Object Metadata Management

Replication Management

Networked Service Application:1/10 G E-Net Management Application Protocol Handling

The Schooner Operating EnvironmentOptimizing Parallel Flash Memory and Multi-Core Processors

22

Schooner Appliance Overview

Schooner Data Access Appliances Purpose-built for Web 2.0 and cloud computing datacenters

8x performance improvements

1/8th the power and space requirements

60% lower TCO

100% compatible with existing applications and management tools

Manufactured, sold and supported worldwide by IBM

Initial Products in the Schooner Appliance Family Schooner Appliance for Memcached Schooner Appliance for MySQL Enterprise™

25

Caching Tier:

Schooner Appliance for Memcached

26

Schooner Appliance for Memcached• Features

– Instantaneous and transparent data persistence, replication, and recoverability

– Multiple containers, allowing separate Memcached domains on a single appliance

– Cache mode or persistent key/value store mode for each container

– Plug-and-play auto configuring

– 100% compatible with existing Memcached applications and monitoring tools

• Benefits– 8x performance improvement over

traditional Memcached servers

– Replaces ordinary servers at a ratio of up to 8:1, yielding both capex and opex savings

Th

rou

gh

pu

t (K

tra

ns/

sec)

Feature OrdinaryServers

Schooner Appliance for Memcached

Containers

Persistence and recoverability

Replication and auto failover

Plug-and-play auto-configuration

GUI and CLI

45,000 TPS

360,000 TPS

© 2009 Schooner Information Technology. All Rights Reserved

27

Legacy Schooner

Typical savings for a mid-size 1TB Web 2.0 datacenter

CapEx:

33 Servers

OpEx

3yr TCO:

$619,000

3yr TCO:

$302,000

OpEx

CapEx: 4

Appliances

Memcached Appliance TCO Analysis

The Bottom Line:

Immediate capex savings

51% TCO savings ($317,000) over three years.

Power and space reductions enable green datacenter initiatives.

With Schooner, consolidation and higher operational efficiency cuts TCO for a rapidly growing 1TB Memcached workload.

Cach

ing

Tie

r

33 2U servers, 18.7 kW (and growing)

Without Schooner

4 2U Schooner appliances, 2.5 kW(Schooner has 8x the capacity of legacy servers,

and can sustain 8x the throughput.)

With Schooner


28

Database Tier:

Schooner Appliance for MySQL Enterprise™

29

Schooner Appliance for MySQL Enterprise™

• Features– Optimized for OLTP (read-write

intensive workloads), as well as OLAP (read mostly workloads)

– Multi-core, high-capacity storage enables scalability and reduces or eliminates sharding

– Fast recovery and warm-up after restart or failover

– Built-in tools enable failover and one-click replication/recovery

– Fully compatible with existing client applications and monitoring tools

– Incorporates highly optimized InnoDB 1.0.3

• Benefits– 8x performance improvement over

legacy disk storage

– Replaces ordinary severs at a ratio of up to 8:1, yielding both capex and opex savings T

ran

sact

ion

s p

er

Min

ute

Run on a typical 2U server

(dual-socket quad-core processor)

Feature OrdinaryServers

Schooner Appliance for MySQL

512 GB flash memory

Reduces or eliminates sharding

Fast recovery and warm-up

Auto failover and one-click replication and recovery

Optimized InnoDB 1.0.3

7,000 TPM

63,000 TPM


30

MySQL Appliance: TCO Analysis

Data

base

Tie

r 20 2U servers, 14.7 kW (and growing)

Without Schooner

2 2U Schooner appliances, 1.4 kW(Schooner has 8x the capacity of legacy servers,

and can sustain 8x the throughput.)

With Schooner

With Schooner, consolidation and higher operational efficiency cuts TCO for a rapidly growing 1TB MySQLworkload.

CapEx:

20 Servers

OpEx

3yr TCO: $563,000

3yr TCO: $197,000

OpEx

CapEx:

2 Schooner Appliances

Typical savings for a mid-size 1TB Web 2.0 datacenter

The Bottom Line:

Immediate capex savings

62% TCO savings ($366,000) over three years

Power and space reductions enable green datacenter initiatives


31

Intel SLC Flash Is a Key Enterprise Technology

READ B/W WRITE B/W ERASE LAT. READ LAT. COST PER GB

HDD 100 mb/s 150.00 mb/s 5,000.00 us $0.10

NAND MLC 250 mb/s 70.00 mb/s 3.5 ms 85.00 us $3.50

NAND SLC 250 mb/s 170.00 mb/s 1.5 ms 75.00 us $11.00

NOR SLC 58 mb/s 0.13 mb/s 5,000.00 ms 0.27 us $70.00

DRAM 2,000 mb/s 2,000.00 mb/s 0.08 us $75.00

Intel’s NAND ® SLC solution - the Intel® X25-E Extreme SATA SSD© 2009 Schooner Information Technology. All Rights Reserved

Limited Benefits of PCI-E or Host-Based Solutions

MetricFusion-io*

io-driveIntel® X25-E SSD 8x Array Unit Note

Read latency uS 4KB 75 75 uSEC

Write Latency uS 4KB 250 85 uSEC Write latency of hardware

Read CPU uS 50 15 uSEC

Write CPU uS 50-100 15 uSECIncreases with garbage collection

Read MBPS 16KB 650 1400 MBPS

Write MBPS 16KB 150 260 MBPS Sustained with garbage collection

Read IOPS 4K 116,000 230,000 IOPS

Write IOPS 4K 20,000 50,000 IOPS Sustained with garbage collection

Capacity 160 512 GB

Cost $7,200 $5,800 USD

Cost/Gbyte $56.25 $11.33 USD* Write throughput assuming continuous random writes spread across the drive, with 20% reserve capacity configured

*Other names and brands may be claimed as the property of others.

32 © 2009 Schooner Information Technology. All Rights Reserved

Schooner SSD Wear Characteristics

Notes: RAID5 spreads access uniformly across SSDs. Schooner appliance for Memcached also uses hash map and multiple internal

volumes to distribute load and wear. FIFO is a write-optimizing policy that performs group writes to SSD.

Innodb - DBT28 Drives

64 GB/drive45,000 TPM

750 Neworders/second15 Buffer writes/new order

4096 Bytes/page11250 Writes/sec per node1406 Writes/sec per drive5.5 MB/Ssec/drive

11930 Seconds to rewrite drive3.3 Hours to rewrite drive3.8 Write amplification0.9 Page reprogram hours

87211 Hours lifetime10.0 Year lifetime

Memcache 5% SET Slab (Typical)8 Drives

64 GB/drive240,000 req/Sec

5% SET fraction1024 Bytes/object

12000 SET/Sec per node1500 Writes/Sec per drive1.5 MB/Sec/drive

44739 seconds to rewrite drive12.4 hours to rewrite drive3.8 Write amplification3.3 page reprogram hours


Memcache 50% SET Slab8 Drives

64 GB/drive100,000 req/Sec


50000 SET/Sec per node6250 Writes/Sec per drive6.1 MB/Sec/drive

10737 seconds to rewrite drive3.0 hours to rewrite drive3.8 Write amplification0.8 page reprogram hours


Memcache 50% SET FIFO8 Drives

64 GB/drive150,000 Req/sec


75000 SET/sec per node9375 Writes/sec per drive9.2 MB/sec/drive

7158 Seconds to rewrite drive2.0 Hours to rewrite drive1 Write amplification

2.0 Page reprogram hours22.7 Year lifetime

In typical and worst-case wear scenarios, life expectancy for SSDs in Schooner appliances ranges from 10 to 30 years:

33


34

Summary: Schooner+Intel+IBMKey Advantages Significantly Lower TCO: Schooner appliances

replace traditional servers by a ratio of up to 8:1, providing immediate capex savings and 60% opex savings

Higher Performance: Schooner appliances provideup to 8x higher performance than traditional servers

Seamless Compatibility: Schooner appliances are 100% compatible with existing client applications and management tools

Quick Deployment: Schooner appliances provide plug-and-play installation for easy set-up and deployment

Easy Management: The Schooner Administrator employs extensive monitoring and optimization features which are easily integrated with existing management tools

Higher Reliability: Schooner appliances employ persistence, replication, and recovery software to deliver enterprise-class reliability and dramatically increase mean time between failures (MTBF)

World-Class Support: IBM provides global, 24/7/365, single-point-of-contact service and support for every Schooner appliance

More Revenue: In addition to improving existing datacenter operations, Schooner appliances support new revenue-producing applications enabled by fast access to terabyte-scale data


35

Thank you.

36

Want More Info on SSDs?

• Attend or download these SSD-related sessionsTuesday, Sept 22nd

– EBLS001 - Extending Battery Life of Mobile PCs: An OverviewWednesday, Sept 23rd

– MEMS001 - Designing Solid-State Drives into Data Center Solutions– MEMS002 - Understanding the Performance of Solid-State Drives in the Enterprise– MEMS003 - Enterprise Data Integrity and Increasing the Endurance of Your Solid-

State Drive– MEMS004 - Future Solid-State Drive Innovations– MEMQ002 - Open Q&A for SSD sessionsThursday, Sept 24th

– MPTS006 - Extreme Notebook Design: Architecting the Most Powerful Mobile Platforms for Gaming & Workstation Applications

– RESS006 - Differentiated Storage Services: Making the Most of Solid-State Drives– STOS004 - Intel® Modular Server with Intel® Solid-State Drives

• Visit our Booth #532 on Level 1 of the Tech Showcase – SSD vs HDD comparisons, gaming demo and more!

• Visit us online at www.intel.com/go/ssd– Product briefs, datasheets, whitepapers, videos, technical support

http://www.intel.com/go/ssd�

37

Legal Disclaimer• INFORMATION IN THIS DOCUMENT IS PROVIDED IN CONNECTION WITH INTEL® PRODUCTS. NO

LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. EXCEPT AS PROVIDED IN INTEL’S TERMS AND CONDITIONS OF SALE FOR SUCH PRODUCTS, INTEL ASSUMES NO LIABILITY WHATSOEVER, AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO SALE AND/OR USE OF INTEL® PRODUCTS INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT. INTEL PRODUCTS ARE NOT INTENDED FOR USE IN MEDICAL, LIFE SAVING, OR LIFE SUSTAINING APPLICATIONS.

• Intel may make changes to specifications and product descriptions at any time, without notice.• All products, dates, and figures specified are preliminary based on current expectations, and are subject to

change without notice.• Intel, processors, chipsets, and desktop boards may contain design defects or errors known as errata, which

may cause the product to deviate from published specifications. Current characterized errata are available on request.

• Performance tests and ratings are measured using specific computer systems and/or components and reflect the approximate performance of Intel products as measured by those tests. Any difference in system hardware or software design or configuration may affect actual performance.

• Intel, Xeon and the Intel logo are trademarks of Intel Corporation in the United States and other countries. • *Other names and brands may be claimed as the property of others.• Copyright © 2009 Intel Corporation.

Risk FactorsThe above statements and any others in this document that refer to plans and expectations for the third quarter, the year and the future are forward-looking statements that involve a number of risks and uncertainties. Many factors could affect Intel’s actual results, and variances from Intel’s current expectations regarding such factors could cause actual results to differ materially from those expressed in these forward-looking statements. Intel presently considers the following to be the important factors that could cause actual results to differ materially from the corporation’s expectations. Ongoing uncertainty in global economic conditions pose a risk to the overall economy as consumers and businesses may defer purchases in response to tighter credit and negative financial news, which could negatively affect product demand and other related matters. Consequently, demand could be different from Intel'sexpectations due to factors including changes in business and economic conditions, including conditions in the credit market that could affect consumer confidence; customer acceptance of Intel’s and competitors’ products; changes in customer order patterns including order cancellations; and changes in the level of inventory at customers. Intel operates in intensely competitive industries that are characterized by a high percentage of costs that are fixed or difficult to reduce in the short term and product demand that is highly variable and difficult to forecast. Additionally, Intel is in the process of transitioning to its next generation of products on 32nm process technology, and there could be execution issues associated with these changes, including product defects and errata along with lower than anticipated manufacturing yields. Revenue and the gross margin percentage are affected by the timing of new Intel product introductions and the demand for and market acceptance of Intel's products; actions taken by Intel's competitors, including product offerings and introductions, marketing programs and pricing pressures and Intel’s response to such actions; and Intel’s ability to respond quickly to technological developments and to incorporate new features into its products. The gross margin percentage could vary significantly from expectations based on changes in revenue levels; capacity utilization; start-up costs, including costs associated with the new 32nm process technology; variations in inventory valuation, including variations related to the timing of qualifying products for sale; excess or obsolete inventory; product mix and pricing; manufacturing yields; changes in unit costs; impairments of long-lived assets, including manufacturing, assembly/test and intangible assets; and the timing and execution of the manufacturing ramp and associated costs. Expenses, particularly certain marketing and compensation expenses, as well as restructuring and asset impairment charges, vary depending on the level of demand for Intel's products and the level of revenue and profits. The current financial stress affecting the banking system and financial markets and the going concern threats to investment banks and other financial institutions have resulted in a tightening in the credit markets, a reduced level of liquidity in many financial markets, and heightened volatility in fixed income, credit and equity markets. There could be a number of follow-on effects from the credit crisis on Intel’s business, including insolvency of key suppliers resulting in product delays; inability of customers to obtain credit to finance purchases of our products and/or customer insolvencies; counterparty failures negatively impacting our treasury operations; increased expense or inability to obtain short-term financing of Intel’s operations from the issuance of commercial paper; and increased impairments from the inability of investee companies to obtain financing. The majority of our non-marketable equity investment portfolio balance is concentrated in companies in the flash memory market segment, and declines in this market segment or changes in management’s plans with respect to our investments in this market segment could result in significant impairment charges, impacting restructuring charges as well as gains/losses on equity investments and interest and other. Intel's results could be impacted by adverse economic, social, political and physical/infrastructure conditions in countries where Intel, its customers or its suppliers operate, including military conflict and other security risks, natural disasters, infrastructure disruptions, health concerns and fluctuations in currency exchange rates. Intel's results could be affected by adverse effects associated with product defects and errata (deviations from published specifications), and by litigation or regulatory matters involving intellectual property, stockholder, consumer, antitrust and other issues, such as the litigation and regulatory matters described in Intel's SEC reports. A detailed discussion of these and other risk factors that could affect Intel’s results is included in Intel’s SEC filings, including the report on Form 10-Q for the quarter ended June 27, 2009.

Rev. 8/3/09

understanding performance of ssds in the enterprise...understanding performance of ssds in the...

Documents