Flash Storage: Trust, But Verify A better, vendor-independent way to analyze flash performance Leah Schoeb, Evaluator Group Peter Murray, Load DynamiX 1
Introductions 2
Speakers Leah Schoeb Senior Partner at Evaluator Group Leah Schoeb brings expertise ranging from cloud infrastructure and virtualization to system performance. Over 20 years of experience in the computer industry with performance engineering and optimization, market positioning, benchmark evidence creation, and guiding industry standards development for server, virtualization, and storage solutions. Peter Murray Senior Product Specialist at Load DynamiX Expert with more than 25 years of experience in storage and network testing. Peter has been extensively involved with the architecture and design of networked testing products for both vendors and global 2000 customers. 3
Agenda Why All Flash Arrays (AFAs) Are Different Challenges in Measuring Modern Flash Array Performance A New Methodology for Measuring AFA Performance The Load DynamiX AFA Performance Validation Solution 4
Why All Flash Arrays are Different Peter Murray, Load DynamiX 5
Landscape for Solid State Storage Greater Performance HPC Data Warehouse Web Applications Transaction Processing Virtualized Applications & VDI Online Archive User Files More Capacity Less Capacity 6/13/2014 6 6
Solid State Storage (Flash) Arrays SSD SSD Flash Module Flash DIMM (or SSD) Traditional Storage Systems Disk replace Hybrid Dynamic Tiering Re-Architect Controller Disk Replace Dynamic Tiering Flash Optimized Controller Disk Replace Dynamic Tiering Architected for Flash Virtual Tiering 2X 5X 5X-10x 10X 6/13/2014 7 7
Considerations for Solid State Storage Arrays Data Services Management Data Reduction Deduplication Compression Thin Provisioning Replication Local (writable) Remote (Future) Management Non-disruptive upgrades REST APIs Investment Protection Self-healing techniques (Reliability) Hardware Redundancy (Availability) Serviceability Support Hypervisor VMware vsphere MS Hyper V Scale out and Clustering Application & OS 8 8
Measuring Accurate Performance All Flash Arrays The Problem Traditional I/O load generation tools don t work Flash has a unique behavior Built-in data services Inline data reduction technologies Different performance curve Very different from hard disk drives Flash arrays perform very differently than traditional HDD systems 6/13/2014 9 9
Inadequate Tools to Validate Flash Storage Performance Measuring new technology based on old assumptions Don t Do It! Result Inflated performance results, inaccurate measurements Causes poor purchasing and deployment decisions The key is setting accurate expectations 6/13/2014 10 10
Modern AFAs Are Different from HDD Arrays Modern Flash Arrays Traditional HDD Arrays Wear leveling Garbage collection Metadata management Self-healing techniques Inline data deduplication Inline compression Thin provisioning Rotational latency Seek times Mechanical parts Controllers designed to handle HDD 6/13/2014 11 11
Measuring with Inline Data Reduction Data content patterns and data streams Patterns written to disk as part of pre-conditioning Patterns presented to an array during steady state Repeating and non-repeating Random patterns Compressible patterns Varying pattern lengths Most I/O load generators are inadequate <.ËT#(âÝ.Èeª..ñn.ä2Õ.Šx7žv.x GöÃc;.¼Â<.ËT#(âÝ.Èeª..ñn.ä2Õ.Šx7žv.x GöÃc;.¼Â<.ËT#(âÝ.Èeª..ñn.ä2Õ.Šx Repeatable noncompressible pattern Repeatable noncompressible pattern Repeatable noncompressible pattern 6/13/2014 12 12
Traditional Disk Performance Curve P E R F O R M A N C E Ramp up Steady State TIME 6/13/2014 13 13
Many Flash Vendors Wide performance variations Pre-conditioning Write Cliff 6/13/2014 14 (SNIA SSSI Specification) 14
Flash Storage Methodology White Paper & SNIA Submission Pending We are submitting this new performance methodology to SNIA SSSI White paper providing more details is available at www.evaluatorgroup.com 6/13/2014 Footer Text 15
New Performance Measurement Solutions are Required for All Flash Arrays How can we set correct expectations? What modern tool set can be used for measuring performance? How can vendors and test sponsors that provide inaccurate results be held accountable? 6/13/2014 16 16
About Load DynamiX 6/13/2014 Footer Text 17
Who is Load DynamiX? Who we are Mission History Headquarters Products Leader in Storage Performance Validation Provide insight into storage infrastructure behavior to assure performance & optimize cost Market entry March 2009; Privately-held Santa Clara, CA Software: Load DynamiX Enterprise Hardware: LDX Series load generation appliances 18
How Load DynamiX Works 4 Steps to Validate Storage Performance Workload Characterization Performance Analytics Workload Modeling Workload Emulation AFA 19
Load DynamiX Helps You Evaluate New Storage Technologies Load DynamiX Validation Appliance Any protocol Any application mix Load DynamiX Workload Modeling GUI FC, iscsi NFS, SMB Object/HTTP Switch AFA AFA Solution A Solution B 20
Methodology In Action Actual results comparing 2 leading AFAs IOPS Comparison for 3 Groups of Data Patterns & R/W Ratios Winner? Depends on your workload. 300000 250000 0% Reducible 50% Reducible 100% Reducible IOPS 200000 150000 100000 Vendor A Vendor B 50000 0 20% / 80% 50% / 50% 80% / 20% 20% / 80% 50% / 50% 80% / 20% 20% / 80% 50% / 50% 80% / 20% Read/Write Ratios 21
Load DynamiX Advantages Highly realistic and consistent modeling Data patterns for complex content generation Unbeatable performance Scales to emulate the largest production workloads Unmatched usability Hides test configuration complexity, no scripting required Small footprint Single 2RU appliance for block, file and object testing vs. 22
Summary Flash has disrupted the storage industry Creates new challenges in measuring performance New testing methodologies are required Load DynamiX solves these challenges Evaluator Group Learn more! www.evaluatorgroup.com Load DynamiX See a product demo! www.loaddynamix.com 23
Q&A Session Leah Schoeb Senior Partner Evaluator Group Peter Murray Senior Product Specialist Load DynamiX 24