FS1 - DOAG
Transcription
FS1 - DOAG
Safe Harbor Statement The following is intended to outline our general product direction. It is intended for information purposes only, and may not be incorporated into any contract. It is not a commitment to deliver any material, code, or functionality, and should not be relied upon in making purchasing decisions. The development, release, and timing of any features or functionality described for Oracle’s products remains at the sole discretion of Oracle. Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 2 Flash und Disk Storage Die Mischung macht’s Franz Haberhauer Chief Technologist Systems Sales Consulting Europe North Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Cost-Performance of Storage Technology Order of magnitude difference must be exploited to optimize solution 12 You cannot afford Flash if you don’t need the performance. Cap HDD 10 0,25, $/IOP =10,00 $/IOP 8 6 Perf HDD 4 27X $IOP You cannot find a better technology than Flash if you need performance. 1,00, $/IOP =3,00 Cap SSD 2 Perf SSD 4,12, $/IOP =0,31 7,50, $/IOP =0,13 0 0 1 2 3 As of January 2014, List prices, approximate Net values 4 $/GB 5 6 Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 7 8 4 Oracle Flash Storage System FS1 Why Multiple Flash Tiers? Mixed Use User Capacity 800GB 400GB Write Intensive 200GB 400GB 200GB Read Intensive 1.6TB 800GB Technology SLC SLC eMLC Over Provisioning 20% 70% 20% 400GB 100% random read 4KB, QD 16 (IOPs) 46,900 43,100 40,400 48,400 47,100 40,000 32,000 34,000 100% random write 4KB, QD 16 (IOPs) 11,300 10,300 7,800 23,200 21,400 2,900 2,700 2,600 100% seq read 1MB, QD 16 (MB/s) 400 330 320 430 420 430 340 340 100% seq write 1MB, QD 16 (MB/s) 210 200 190 270 270 80 70 60 Full Drive Writes Per Day 10 Price/GB ~2.5 Warranty 5 Years or Max TBW Performance Gated <3 1 5 Years 3 Years or Max TBW Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal and NDA only 5 Oracle Flash Storage System FS1 Why Multiple Flash Tiers? Mixed Use User Capacity 800GB 400GB Write Intensive 200GB 400GB 200GB Read Intensive 1.6TB 800GB Technology SLC SLC eMLC Over Provisioning 20% 70% 20% 400GB 100% random read 4KB, QD 16 (IOPs) 46,900 43,100 40,400 48,400 47,100 40,000 32,000 34,000 100% random write 4KB, QD 16 (IOPs) 11,300 10,300 7,800 23,200 21,400 2,900 2,700 2,600 100% seq read 1MB, QD 16 (MB/s) 400 330 320 430 420 430 340 340 100% seq write 1MB, QD 16 (MB/s) 210 200 190 270 270 80 70 60 Full Drive Writes Per Day 10 Price/GB ~2.5 Warranty 5 Years or Max TBW Performance Gated <3 1 5 Years 3 Years or Max TBW Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal and NDA only 6 Access Skew and Data Access Patterns Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal/Restricted/Highly Restricted 7 Access Skew Data Resident In a Storage System • Only ~5% of data is IO-active at any moment in time • But the particular 5% changes dynamically based on business process cycles, market events, “black swans” • The ~5% active data isn’t always, or necessarily, the most important data for business Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal/Restricted/Highly Restricted 8 Data Skew The new important metric in storage • ~ 95% cold – does not need a lot of IOPS • Put hot data on expensive flash and cold on inexpensive capacity media. • Problem: data temperature constantly changing – As hot data ages it typically cools • in some cases tends to heat up again – The trick is to both size the total amount of flash you will need and to figure out what data is hot and place it on flash • Then factor all decisions by the business value of the data Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 9 OLTP Transactional Data (E-Business Suite) The most common use case. • Hot during the transaction, • warm during the billing month (shipping, restocking, store dailies, etc.) • cold (read only) until monthly/quarterly/annual reports are run. – after this data stone cold, but can’t delete it, • store it on the least expensive media possible. • Data is hot or warm for about 30 days (monthly billing period), – so over a year only 8.3% of your data is hot. • this is probably an over estimate • average billing latency is 15 days, a better number might be 4.15% or something in between Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 10 Financial Services: Demand Deposit Accounting (FlexCube) Checks, ATM transactions, PayPal etc. • Hot from transaction initiation, daily posting and balancing warm through monthly statement and billing runs and payment processing. – 15-30 days hot to very warm, 30-45 days warm, >45 days cold. – Somewhat similar to pure OLTP type transactions • Once again cannot then just delete this data. 7 year retention is a regulatory minimum. • Summary: 4.15% to 8.3% of the data is read/write hot, about 2-3X that is sequential read hot with quarterly reporting. – may want to consider different flash media types here… Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 11 Telco Call Detail Records (CDR’s) remediation and billing • Make a call, generate a CDR and store it for an average of 15 days till the next billing cycle. • Remediate the CDR’s so all parties to the call get paid their percentage • Sort the CDR’s by phone number for billing and archival purposes • Occasionally extract CDR’s for some govt. agencies. Then archive them. • Again typical 30 day (average 15 day latency) cycle, 4.15 to 8.3% of the data hot in a read/write mode. Almost all read thereafter. Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 12 Application Workload Skew Example SPC-1 Workload (OLTP Benchmark) Accumulative percent of accesses in FLASH 120 Cum Accesses % 100 SPC-1 Workload 5.5% of data gets 90.6% of IO’s 80 Pure Random Workload 60 Skew = 0% 40 20 0 0 10 20 30 40 50 60 70 % of Total of Capacity for Application Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 80 90 100 Lessons learned There’s more to storage than IOPS and capacity • Hot data tends to be in the 4-6% range based on a years worth of data. – If longer periods are modeled, the hot percentage will go down. A little Flash goes a long way! • Hot data is generally read/write, becomes sequential read only as it cools. • When thinking about auto-tiering with flash: – eMLC flash is great at reading and not so good at writing, but it costs a lot less than SLC flash. – SLC flash is good at both, but costs more. – eMLC flash also has less write endurance than SLC so it should not be used in write heavy environments. Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 14 Lessons learned There’s more to storage than IOPS and capacity • Hot data tends to be in the 4-6% range based on a years worth of data. – If longer periods are modeled, the hot percentage will go down. A little Flash goes a long way! • Hot data is generally read/write, becomes sequential read only as it cools. Multiple flash tiers • When thinking about auto-tiering with flash: – eMLC flash is great at reading and not so good at writing, but it costs a lot less than SLC flash. – SLC flash is good at both, but costs more. – eMLC flash also has less write endurance than SLC so it should not be used in write heavy environments. Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 15 Data Skew and the Law of diminishing Returns • I need 25,000 IOPS and I have 50 TB of data – How much Flash should I buy? • Coming up. – Is there a “Flash Analysis Utility” • No, excepting hokus-pokus marketing web tools. Useless. – When does my data transition from read/write to read? • • • • Generally after 30 days. Perhaps I should put all my data on flash as the All Flash Array vendors suggest? With 95% of your data generally cold, why would you do this? And they claim super-duper effective deduplication, so the $/TB is equal to HDD’s! Really? Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 16 Diminishing Returns When to stop buying flash! A Rule of Thumb: Sum of the % of IOPS and % of capacity = 1 A Little Flash Goes a Long Way! Diminishing Return Point 100% + 100% = 2 100 90 80 85% of IOPS + 15% of Capacity = 1 IOPS 70 60 50 40 30 20 10 0 0 0% + 0% = 0 10 High marginal Value 20 30 40 50 60 70 80 90 100 Capacity Diminishing marginal Value Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 17 Lessons learned 2 • Don’t over buy flash. – Start with 4-6% of your application data total capacity – Add some margin for expected data growth – You can always add more later. Overbuying follows the law of diminishing returns • The Oracle FS1 can model un-installed flash to tell you how it would use it if it was installed. • Look for special Cases – VDI Boot Storms: Pin Golden Images to Flash. Just add them up. – Data Warehouse/Business Intelligence Analytics • Consider Exadata + Exalytics. Otherwise consider server-side flash with Oracle 12c & ADO. – Certain simulation data are all “hot”: (Fluent, Crash, NASTRAN). Pin data in flash. • Input and output is sequential, does not require flash. Flash for scratch only Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 18 Oracle: An Established Flash Foundation Over Half a Decade of Flash Technology Leadership 12.8TB L2ARC Logzilla 2009 Exadata 2008 ZFS Copyright © 2014, Oracle and/or its affiliates. All rights reserved. 4TB FMod Flash DIMMs 2009 Database 11gR2 44.8 TB Smart Flash Cache im Fullrack 2009 Axiom 19 2009 F5100 2009 F Series PCIe Flash Cards Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | F80 800GB Introducing the Oracle FS1 Flash Storage System • The most intelligent flash storage system • Designed from the onset to maximize the power of flash and the economics of disk • Adaptive and anticipatory storage with rapid learning and highly granular data tiering • Co-Engineered for Oracle on Oracle Advantage Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 20 FS1: Advanced High Availability Scale-Out Architecture Grows with Your Business • Up to 16 HA Nodes • Petabytes of Flash • Up to 2M 50/50 R-W IOPS • Up to 80 GB/sec Throughput • Full Suite of Data Services Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 21 Oracle FS1 Storage System • Hardware – Controllers • Standard Oracle Servers • NVDIMMs: Flash-Backed RAM Cache • Up to 8 Replication Engines – Drive Enclosures • Up to 30 DE’s per FS1-2 • Common with ZS – Drives • 400GB SLC & 1.6TB eMLC SSD/Flash • 300GB, 900GB, and 4TB SAS-2 HDD – Rack • Common look and feel Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal/Restricted/Highly Restricted 22 FS1-2 Disk Enclosures SSD Enclosures Performance • 400GB • 7 pop, 2.8TB •400GB • 13 pop, 5.2TB HDD Enclosures Capacity Performance • 1600GB • 7 pop, 11.2TB • SAS 10k 300GB • 24 pop, 7.2TB •1600GB • 13 pop, 20.8TB Capacity • SAS 7.2k 4TB • 24 pop, 96TB • SAS 10k 900GB • 24 pop, 21.6TB •1600GB • 19 pop, 30.4TB up to 228 TB Performance or 912 TB Capacity Flash 2.9 PB HDD Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal/Restricted/Highly Restricted 25 Storage QoS Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal/Restricted/Highly Restricted 26 FS1 QoS Plus – Intelligent Storage Management QoS Plus Adapts to Lowest Cost & Highest Performance by Business Priority QoS Plus: a policy-based virtualization feature incorporating business priority and performance optimization fused with sub-LUN automatic tiering into one simple management framework. Automatically. Performance Flash Capacity Flash Continuous Pin Learn & Hold Adjust Rate Throttle $/IOP Performance Disk Control Modes Capacity Disk $/TB Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential - Restricted 27 Operational Modes Not all data or Application Environments are the same • Not just Off or On • Five basic modes: 1. Fast Learn: Execute data migration operation three times during the first period, then once per period thereafter 2. Hold: Immediately stop data migration, but keep learning 3. Adjustable Scan: User specifies time period for scan and data migration. • • Default is 24 hours Learning is continuous. 4. Throttle Movement: Hold, Fast, Slow or System Chooses 5. Pin: Do not move this data regardless of use or QoS setting Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Operational Modes: Span of Control Coarse to Granular • User can set “global” QoS Plus settings within a Storage Domain – Defaults to all LUNS within a Domain • Specific LUNS may have different “override” settings based on application or user requirements • e.g. Exclude a specific media type from QoS Plus migration • Shadow Learn: Learn as if a not installed tier was present • Great tool to evaluate the effect of adding a new tier • Settings may be changed at any time Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Access Skew: What is HOT and What is NOT FS1 QoS Plus divides data into fixed 640k fine grained “blocks” to build a ‘Heat Map’ Hot – Frequent access Warm– semi- frequent access Cold– infrequent access Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal/Restricted/Highly Restricted 30 What is HOT and What is NOT changes with time QoS Plus continuously maps changing access patterns Hot – Frequent access Warm– semi- frequent access Cold– infrequent access Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal/Restricted/Highly Restricted 31 QoS Plus Auto-Tiering Granular Efficiency FS1 3PAR VNX2 EMC 640K 256 MB (400X an FS1) Bigger is not better 1 GB (VNX1) (1600X an FS1) Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal/Restricted/Highly Restricted 32 Bigger isn‘t Better 400 Oracle Pixels 1 256MB Chunklet • 25 MB in Perf SSD • 0 MB in SSD • 256 MB in SSD • 100 MB in Perf Disk • 0 MB in Perf Disk • 0 MB in Perf Disk • 131 MB in Cap Disk • 256 MB in Cap Disk • 0 MB in Cap Disk Performance: Performance: Performance: IOPS = 81.1K IOPS = 1K IOPS = 90K Latency = 1.4 mS Latency = 6 mS Latency = 0.5mS Cost = $0.250 Cost = $0.064 Cost = $1.92 Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal and NDA only 33 Bigger isn‘t Better 400 Oracle Pixels 1 256MB Chunklet Bigger results in 10% more IOPS for 7X the cost!! • 25 MB in Perf SSD • 100 MB in Perf Disk • 131 MB in Cap Disk • 0 MB in SSD OR• • 256 MB in SSD 0 MB in Perf Disk • 0 MB in Perf Disk • 256 MB in Cap Disk • 0 MB in Cap Disk 1/4th 1.2% of the IOPS for the cost. Performance: Performance: Performance: IOPS = 81.1K IOPS = 1K IOPS = 90K Latency = 1.4 mS Latency = 6 mS Latency = 0.5mS Cost = $0.250 Cost = $0.064 Cost = $1.92 Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal and NDA only 34 Intelligent Management. Automatically. Intelligence Beyond Access Frequency Real Time Enterprise COLLECT COLLECT STATISTICS Read, Write, Random , Sequential, Frequency, and Priority counters for each 640K sub-lun “block” EVALUATE QoS Plus MOVE • Most granular data “blocks” EVALUATE THE STATISTICS Analyze counters to construct list for promotion/demotion move candidates MOVE DATA Kickoff a background process to promote or demote the 640K sub-lun “block” Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | • Comprehensive access profile collection, not just access frequency • Fast Learn Mode • Adjustable Scanning Periods • Discrete migration options and controls Oracle Confidential – Internal/Restricted/Highly Restricted 35 IO Management with Oracle FS1 QoS Plus: IO Prioritization Based on Business Value I/O Requests to Storage Systems: 10 9 8 7 6 5 4 3 2 1 FS1 Priority-In-Priority-Out Queue: 10 9 8 7 6 5 4 3 2 1 Premium Priority High Priority Medium Priority Low Priority Archive Priority FS1 Intelligent IO Management Manage IOs to Business Priorities Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal/Restricted/Highly Restricted 37 Normalized Access Frequency Profiles QoS Plus: QoS fused with Auto-Tiering 100% 80% 60% Perf SSD 40% Cap SSD 20% Perf HDD 0% Cap HDD QoS Level: Business Priority Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Oracle Confidential – Internal/Restricted/Highly Restricted 38 Oracle FS1—the Most Intelligent Storage Provisioning Co-Engineered, Tuned, and Tested for Oracle Database and Many Other Apps Provision Storage for An Oracle Database? 1-click 7 Oracle Database Components Auto-provisioned and Mapped to Optimum Storage Tiers mnt/dbname/idx (random/read) mnt/dbname/control (mixed, mixed) mnt/dbname/archivelog (seq, write) mnt/dbname/redo (seq, write) mnt/dbname/tempfile (mixed/mixed) mnt/dbname/datafile (mixed/mixed) mnt/dbname/chgtrack (seq, write) Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 40 ... Oracle FS1 Application Profiles One-Click Provisioning with Proven Best-Practices: 1 2 3 4 An administrator wants to provision storage for an application Launches the FS Storage System Manager Selects the right application profile Storage is provisioned at a sub-application, file level based on storage class and priority of the data as set by the application profile Application profiles are pre-defined and tested for optimal provisioning: • QoS Level (IO Priority) • Access and IO Bias • RAID level/thin provisioning • Caching behavior and read-ahead mnt/dbname/idx (random/read) mnt/dbname/control (mixed, mixed) mnt/dbname/archivelog (sequential, write) mnt/dbname/redo (sequential, write) mnt/dbname/tempfile (mixed/mixed) mnt/dbname/datafile (mixed/mixed) mnt/dbname/chgtrack (sequential, write) Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 41 Application Profiles • Pre-tested optimal provisioning and QoS settings include –QoS Level (I/O Priority) –Access and IO Bias –RAID level/Thin Provisioning –Caching behavior and read-ahead Oracle Database / OLTP Example Data Type • Most application types ship pre-built with every FS – – – – – Oracle Database / OLTP Archive / ILM Business Analytics / OLAP E-mail: Microsoft Exchange Business Applications: CRM, ERP, HR Recommended Storage Class Priority Access Bias I/O Bias Control Files SSD / perf HDD Premium - High Mixed Mixed Database Index SSD Premium - High Mixed Mixed Database Tables SSD / perf or cap HDD Premium - Medium Mixed Mixed Temp Files Cap HDD Medium Mixed Mixed Online redo Log Files Perf HDD High Sequential Write Archive Log Files Cap HDD Low Sequential Write Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | QoS Plus: Priority Aligned Autonomous Tiering Set QoS by App Workload-Driven Heat Maps Fine-Grain Auto Tiering CPU, Cache, Tier, IOPS, Response Time Archive Priority Low Priority High Priority Medium Priority Secure Multi-Tenant Storage Domains Performance Flash Capacity Flash Premium Priority Performance Disk Capacity Disk Hot Data Warm Data Cold Data Deep Archive 4 Storage Tiers 2 Flash, 2 Disk Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 43 Performance 2 Node HA Building Block, All-Flash, FC FS1 173 292 118 20 176 27 0.6 2.2 365 186 265 215 5,4 10.8 8.6X 1.2X 1.5X 1.5X 8X 9.7X 4.9X EMC EMC EMC EMC EMC EMC EMC …less than half the price per TB EMC performance source http://xtremio.com/wp-content/uploads/2014/09/h13419_White-Paper_XtremIO_Ver_2-4-1_Performance-Report.pdf published September 2014, EMC List Price Source of $510,000 for 20TB X-Brick, source http://www.emc.com/sales/stateoffl/florida-price-list-2014-05.pdf , Oracle performance from internal testing, Maximum Oracle FS1-2 flash capacity is 900TB, 173TB recommended for high-performance deployments. Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 46 High Performance Flash Storage for Enterprise SAN Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | Copyright © 2014, Oracle and/or its affiliates. All rights reserved. | 48