SlideShare a Scribd company logo
Storage media solution
"including pricing as a factor"
Tony Roug
Rougs.com LLC
Storage Technology Consulting
August 15, 2013
Where does pricing fit?
• Storage decision are
driven by
– Capacity
– Performance
– Price
• In a world of many
technologies and usage
models
– what's a simple
visualization of the
above
– propose best price
tradeoff for single
server configs
Rougs.com LLC Copyright
HDDSSDPCIe SSDDDR
or or or
or
What's the low cost
for a Performance/Capacity
requirement
assumes a "commodity" model for storage hardware
Configuration
• Server configuration
– Basic SuperMicro 1U and 2U rack
23.5" server
– processor/board: Xeon E5 C600
Chipset – Single and Dual
– drive slots: 1U - 10, 2U –
24, assume possible for either 2.5
or 3.5 (i.e. HDDs.)
– PCIe slots: 1U – 2, 2U – 4, only
consider half height, half width
– DDR: up to 1TB per server
– All configurations forced to single
server
– Servers Direct Pricing
• SATA/SAS HDDs
– Seagate 15K/10K and 7.2K SATA
– Servers Direct pricing
– IOPS based on rotational speed
• SATA SSDs:
– SATA SSDs (Intel)
– Amazon and NewEgg pricing
– Spec sheet 4K random
performance
• PCIe SSD:
– Intel 910, Amazon Pricing
– No Fusion because no public price
or performance info
• DDR DRAM:
– Kinston
– Servers Direct Pricing
– Spec Sheet performance
• Notes/Assumptions
– IOPS Performance scales linearly as
drives are added
– No extra drives for reliability or
availability
Rougs.com LLC Copyright
Base data for price calculation
Rougs.com LLC Copyright
July 16, 2013
July 8, 2010
In this case all data based on spec sheets and web pricing
Random 4K 90% Read, 10% Write
Workloads: TPC-E
Rougs.com LLC Copyright
100
1,000
10,000
100,000
1,000,000
128 256 512 1024 2048 4096 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD X25M X25M DC X25E DDR
July 16, 2013July 8, 2010
PCIE SSD
Opportunity
PCIe opportunity in 2010 – reduced # drives and performance
100
1,000
10,000
100,000
1,000,000
128 512 2048 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
TPC-E target for
DB size
Random 4K 70% Read, 30% Write
Workloads: TPC-C
Rougs.com LLC Copyright
100
1,000
10,000
100,000
1,000,000
128 256 512 1024 2048 4096 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD X25M X25M DC X25E DDR
TPC-C target for
DB size
July 16, 2013July 8, 2010
PCIE SSD
Opportunity
PCIe card opportunity in 2010 – reduced # drives and performance
100
1,000
10,000
100,000
1,000,000
128 256 512 1024 2048 4096 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
Random 4K 100% Read
workloads: boot/VDI, digital content, Hadoop
Rougs.com LLC Copyright
100
1,000
10,000
100,000
1,000,000
128 512 2048 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD X25M X25M DC X25E DDR
July 16, 2013July 8, 2010
Typical 100%
read workload
No opportunity for PCIe SSDs in 2010 or today
100
1,000
10,000
100,000
1,000,000
128 512 2048 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
Random 4K 100% Write
Workloads: FSI trading, DB Logging
Rougs.com LLC Copyright
July 16, 2013July 8, 2010
100
1,000
10,000
100,000
1,000,000
128 512 2048 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD X25M X25M DC X25E DDR
PCIE SSD
Opportunity
PCIe SSD filled opportunity hole in 2010, performance and reduced number of
drives
100
1,000
10,000
100,000
1,000,000
128 256 512 1024 2048 4096 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
Random 4K 60% Read, 40% Write
Workloads: CDN Web Caching
Rougs.com LLC Copyright
100
1,000
10,000
100,000
1,000,000
128 512 2048 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD X25M X25M DC X25E DDR
July 16, 2013July 8, 2010
Typical workload
area
Opportunity for CDN covered by SSDs in 2010
100
1,000
10,000
100,000
1,000,000
128 512 2048 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
Choosing lowest power versus price
HDDSSDPCIe SSDDDR
or or or
What's the low power
for a Performance/Capacity
requirement
Random 4K 70% Read, 30% Write
Comparing selecting on price versus power
Power - July 16, 2013
100
1,000
10,000
100,000
1,000,000
128 256 512 1024 2048 4096 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
100
1,000
10,000
100,000
1,000,000
128 256 512 1024 2048 4096 8192
PeakIOPS
Total App Gig Required
SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
Power - July 16, 2013
When focused on just power, SSDs almost always beat HDDs and DDR
Summary
• For a simple study on spec price/performance
• Commercial SATA SSDs have significantly improved in
3 years
• Can cover most of the common usage cases
• For typical usage models, performance of SATA
SSD seems largely good enough
• For typical usage models, SATA SSD seem to be
the lowest power for almost all usage models

More Related Content

PPTX
Seagate 7200 vs wd 5400
PDF
robust-twelve-plus-midtower-storage-server
PDF
How to build a state-of-the-art rails cluster
PDF
Ceph Day San Jose - All-Flahs Ceph on NUMA-Balanced Server
PPTX
JetStor JBOD Microsoft Storage Spaces Xces BV
PPTX
Ceph Day Taipei - How ARM Microserver Cluster Performs in Ceph
PDF
Ceph Day Beijing - Welcome to Beijing Ceph Day
PDF
Propelling IoT Innovation with Predictive Analytics
Seagate 7200 vs wd 5400
robust-twelve-plus-midtower-storage-server
How to build a state-of-the-art rails cluster
Ceph Day San Jose - All-Flahs Ceph on NUMA-Balanced Server
JetStor JBOD Microsoft Storage Spaces Xces BV
Ceph Day Taipei - How ARM Microserver Cluster Performs in Ceph
Ceph Day Beijing - Welcome to Beijing Ceph Day
Propelling IoT Innovation with Predictive Analytics

What's hot (15)

PPTX
Ceph Day KL - Ceph on All-Flash Storage
PPTX
Ceph Day KL - Ceph on ARM
PPTX
Ceph Day San Jose - Ceph at Salesforce
PPTX
Ceph Day Seoul - The Anatomy of Ceph I/O
PPTX
MySQL Head-to-Head
PPTX
Application_Benchmark_into_Virtualization
PDF
Ceph Day Seoul - Delivering Cost Effective, High Performance Ceph cluster
PDF
Ceph Day Beijing - Ceph All-Flash Array Design Based on NUMA Architecture
PDF
The new AMD EPYC solutions from OVHcloud: what benefits?
PPTX
Ceph Day KL - Delivering cost-effective, high performance Ceph cluster
PDF
Nimble-Storage-AFA-Datasheet
PDF
Ceph Day Tokyo - Bit-Isle's 3 years footprint with Ceph
PPTX
Ceph Day KL - Ceph Tiering with High Performance Archiecture
PDF
Ceph Day Beijing - Storage Modernization with Intel and Ceph
PDF
Cy7 introduction
Ceph Day KL - Ceph on All-Flash Storage
Ceph Day KL - Ceph on ARM
Ceph Day San Jose - Ceph at Salesforce
Ceph Day Seoul - The Anatomy of Ceph I/O
MySQL Head-to-Head
Application_Benchmark_into_Virtualization
Ceph Day Seoul - Delivering Cost Effective, High Performance Ceph cluster
Ceph Day Beijing - Ceph All-Flash Array Design Based on NUMA Architecture
The new AMD EPYC solutions from OVHcloud: what benefits?
Ceph Day KL - Delivering cost-effective, high performance Ceph cluster
Nimble-Storage-AFA-Datasheet
Ceph Day Tokyo - Bit-Isle's 3 years footprint with Ceph
Ceph Day KL - Ceph Tiering with High Performance Archiecture
Ceph Day Beijing - Storage Modernization with Intel and Ceph
Cy7 introduction
Ad

Similar to A way to visual the best storage media for an application (20)

PPTX
Deploying ssd in the data center 2014
PPT
Open Ware Ramsan Dram Ssd
PDF
Power of Storage FREE Webinar
PDF
Toshiba Storage Protfolio 2014
PPT
Apresentacao Solid Access Corp Presentation Openware 5 20 10
PDF
Database performance tuning for SSD based storage
PDF
Josh Krischer - How to get more for less (4 november 2010 Storage Expo)
PPTX
Flash memory summit enterprise udate 2019
PPTX
2015 deploying flash in the data center
PPTX
2015 deploying flash in the data center
PDF
Marvell FMS Preso
PDF
Infoboom future-storage-aug2011-v3
PDF
Infoboom future-storage-aug2011-v3
PPTX
002-Storage Basics and Application Environments V1.0.pptx
PDF
How to Successfully Deploy SSD in the Enterprise
PDF
Hp smart cache technology c03641668
PDF
High Performance Hardware for Data Analysis
PDF
Mike Pittaro - High Performance Hardware for Data Analysis
PDF
Kinetic basho public
PPTX
How Big Data is Transforming the Data Center
Deploying ssd in the data center 2014
Open Ware Ramsan Dram Ssd
Power of Storage FREE Webinar
Toshiba Storage Protfolio 2014
Apresentacao Solid Access Corp Presentation Openware 5 20 10
Database performance tuning for SSD based storage
Josh Krischer - How to get more for less (4 november 2010 Storage Expo)
Flash memory summit enterprise udate 2019
2015 deploying flash in the data center
2015 deploying flash in the data center
Marvell FMS Preso
Infoboom future-storage-aug2011-v3
Infoboom future-storage-aug2011-v3
002-Storage Basics and Application Environments V1.0.pptx
How to Successfully Deploy SSD in the Enterprise
Hp smart cache technology c03641668
High Performance Hardware for Data Analysis
Mike Pittaro - High Performance Hardware for Data Analysis
Kinetic basho public
How Big Data is Transforming the Data Center
Ad

Recently uploaded (20)

PDF
Mobile App Security Testing_ A Comprehensive Guide.pdf
PDF
GDG Cloud Iasi [PUBLIC] Florian Blaga - Unveiling the Evolution of Cybersecur...
PDF
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
PPTX
Effective Security Operations Center (SOC) A Modern, Strategic, and Threat-In...
PPTX
Big Data Technologies - Introduction.pptx
PDF
Approach and Philosophy of On baking technology
PDF
Dropbox Q2 2025 Financial Results & Investor Presentation
PDF
Electronic commerce courselecture one. Pdf
PDF
Modernizing your data center with Dell and AMD
PDF
How UI/UX Design Impacts User Retention in Mobile Apps.pdf
PPTX
20250228 LYD VKU AI Blended-Learning.pptx
PDF
Reach Out and Touch Someone: Haptics and Empathic Computing
PDF
Per capita expenditure prediction using model stacking based on satellite ima...
PDF
Machine learning based COVID-19 study performance prediction
PDF
NewMind AI Monthly Chronicles - July 2025
PPTX
PA Analog/Digital System: The Backbone of Modern Surveillance and Communication
PDF
Empathic Computing: Creating Shared Understanding
PPTX
breach-and-attack-simulation-cybersecurity-india-chennai-defenderrabbit-2025....
PDF
Unlocking AI with Model Context Protocol (MCP)
PDF
Diabetes mellitus diagnosis method based random forest with bat algorithm
Mobile App Security Testing_ A Comprehensive Guide.pdf
GDG Cloud Iasi [PUBLIC] Florian Blaga - Unveiling the Evolution of Cybersecur...
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
Effective Security Operations Center (SOC) A Modern, Strategic, and Threat-In...
Big Data Technologies - Introduction.pptx
Approach and Philosophy of On baking technology
Dropbox Q2 2025 Financial Results & Investor Presentation
Electronic commerce courselecture one. Pdf
Modernizing your data center with Dell and AMD
How UI/UX Design Impacts User Retention in Mobile Apps.pdf
20250228 LYD VKU AI Blended-Learning.pptx
Reach Out and Touch Someone: Haptics and Empathic Computing
Per capita expenditure prediction using model stacking based on satellite ima...
Machine learning based COVID-19 study performance prediction
NewMind AI Monthly Chronicles - July 2025
PA Analog/Digital System: The Backbone of Modern Surveillance and Communication
Empathic Computing: Creating Shared Understanding
breach-and-attack-simulation-cybersecurity-india-chennai-defenderrabbit-2025....
Unlocking AI with Model Context Protocol (MCP)
Diabetes mellitus diagnosis method based random forest with bat algorithm

A way to visual the best storage media for an application

  • 1. Storage media solution "including pricing as a factor" Tony Roug Rougs.com LLC Storage Technology Consulting August 15, 2013
  • 2. Where does pricing fit? • Storage decision are driven by – Capacity – Performance – Price • In a world of many technologies and usage models – what's a simple visualization of the above – propose best price tradeoff for single server configs Rougs.com LLC Copyright HDDSSDPCIe SSDDDR or or or or What's the low cost for a Performance/Capacity requirement assumes a "commodity" model for storage hardware
  • 3. Configuration • Server configuration – Basic SuperMicro 1U and 2U rack 23.5" server – processor/board: Xeon E5 C600 Chipset – Single and Dual – drive slots: 1U - 10, 2U – 24, assume possible for either 2.5 or 3.5 (i.e. HDDs.) – PCIe slots: 1U – 2, 2U – 4, only consider half height, half width – DDR: up to 1TB per server – All configurations forced to single server – Servers Direct Pricing • SATA/SAS HDDs – Seagate 15K/10K and 7.2K SATA – Servers Direct pricing – IOPS based on rotational speed • SATA SSDs: – SATA SSDs (Intel) – Amazon and NewEgg pricing – Spec sheet 4K random performance • PCIe SSD: – Intel 910, Amazon Pricing – No Fusion because no public price or performance info • DDR DRAM: – Kinston – Servers Direct Pricing – Spec Sheet performance • Notes/Assumptions – IOPS Performance scales linearly as drives are added – No extra drives for reliability or availability Rougs.com LLC Copyright
  • 4. Base data for price calculation Rougs.com LLC Copyright July 16, 2013 July 8, 2010 In this case all data based on spec sheets and web pricing
  • 5. Random 4K 90% Read, 10% Write Workloads: TPC-E Rougs.com LLC Copyright 100 1,000 10,000 100,000 1,000,000 128 256 512 1024 2048 4096 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD X25M X25M DC X25E DDR July 16, 2013July 8, 2010 PCIE SSD Opportunity PCIe opportunity in 2010 – reduced # drives and performance 100 1,000 10,000 100,000 1,000,000 128 512 2048 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR TPC-E target for DB size
  • 6. Random 4K 70% Read, 30% Write Workloads: TPC-C Rougs.com LLC Copyright 100 1,000 10,000 100,000 1,000,000 128 256 512 1024 2048 4096 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD X25M X25M DC X25E DDR TPC-C target for DB size July 16, 2013July 8, 2010 PCIE SSD Opportunity PCIe card opportunity in 2010 – reduced # drives and performance 100 1,000 10,000 100,000 1,000,000 128 256 512 1024 2048 4096 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
  • 7. Random 4K 100% Read workloads: boot/VDI, digital content, Hadoop Rougs.com LLC Copyright 100 1,000 10,000 100,000 1,000,000 128 512 2048 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD X25M X25M DC X25E DDR July 16, 2013July 8, 2010 Typical 100% read workload No opportunity for PCIe SSDs in 2010 or today 100 1,000 10,000 100,000 1,000,000 128 512 2048 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
  • 8. Random 4K 100% Write Workloads: FSI trading, DB Logging Rougs.com LLC Copyright July 16, 2013July 8, 2010 100 1,000 10,000 100,000 1,000,000 128 512 2048 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD X25M X25M DC X25E DDR PCIE SSD Opportunity PCIe SSD filled opportunity hole in 2010, performance and reduced number of drives 100 1,000 10,000 100,000 1,000,000 128 256 512 1024 2048 4096 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
  • 9. Random 4K 60% Read, 40% Write Workloads: CDN Web Caching Rougs.com LLC Copyright 100 1,000 10,000 100,000 1,000,000 128 512 2048 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD X25M X25M DC X25E DDR July 16, 2013July 8, 2010 Typical workload area Opportunity for CDN covered by SSDs in 2010 100 1,000 10,000 100,000 1,000,000 128 512 2048 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR
  • 10. Choosing lowest power versus price HDDSSDPCIe SSDDDR or or or What's the low power for a Performance/Capacity requirement
  • 11. Random 4K 70% Read, 30% Write Comparing selecting on price versus power Power - July 16, 2013 100 1,000 10,000 100,000 1,000,000 128 256 512 1024 2048 4096 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR 100 1,000 10,000 100,000 1,000,000 128 256 512 1024 2048 4096 8192 PeakIOPS Total App Gig Required SATA HDD SAS HDD Intel 3500 Intel 3700 Intel 910 DDR Power - July 16, 2013 When focused on just power, SSDs almost always beat HDDs and DDR
  • 12. Summary • For a simple study on spec price/performance • Commercial SATA SSDs have significantly improved in 3 years • Can cover most of the common usage cases • For typical usage models, performance of SATA SSD seems largely good enough • For typical usage models, SATA SSD seem to be the lowest power for almost all usage models