1
FLASH 1 ST THE STORAGE STRATEGY FOR THE NEXT DECADE Iztok Sitar Sr. Technology Consultant EMC Slovenia 2
Information Tipping Point Ahead The Future Will Be Nothing Like The Past 140,000 120,000 100,000 80,000 60,000 40,000 110X MORE DATA EACH DECADE A 10TB Data Center In 2001 Growing At 60% Y/Y Will Be A 120 Petabyte Data Center By We are Here 2021, - But IT Budgets Remain Flat Data Automation 20,000 Budget 0 2000 2003 2006 2009 2012 2015 2018 2021 3
Most Data Will Be Stone Cold While We Increasingly Can t Erase Data, We Can Store It Better 37.5 TB 4,123 TB 110 X 2011 2021 Hot Warm Cold Evolution Of 50 TBs In 10 Years 4
Why Disk Aggregation Is Losing Steam Moore s Law Drives The Escalating Need For IO Transactions The Number Of IOPS Needed Per Host Over Time? We are Here 23 36 58 93 148 237 378 603 962 1.536 2.451 3.912 6.243 9.964 64.652 40.509 25.381 15.903 103.184 164.682 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 262.833 5
The CPU To HDD Performance Gap CPU Improves 100 Times Every Decade Disk Speed Hasn t MOORE S LAW: 100X per DECADE CPU Continue To Improve While Disk Drive Performance Remains Flat. As A Result, Applications Will Suffer More And More Unless We Rapidly Move To FLASH. 100 times improved FLASH 10,000 times improved 2000 2010 2020 6
Comparing Associated Costs Which Technology Is The Most Efficient? Capacity Acquisition Cost $- $5 $10 $15 $20 $25 $30 Capacity Power Cost 0 10 20 30 15K HDD 7200 HDD FLASH Lowest Capacity Cost 15K HDD 7200 HDD FLASH Transaction Acquisition Cost $- $5 $10 $15 $20 Transaction Power Cost 0,0 50,0 100,0 150,0 15K HDD 7200 HDD FLASH Lowest Transaction Cost 15K HDD 7200 HDD FLASH 7
Two Major Components Of Your SLA Transaction And Capacity Support Needed By The Business? SLA Target SLA Lowest $/IO SSD Needed Transactions $ Lowest $/GB Needed Capacity NL-HDD 8
FLASH Becomes More Affordable Projected To Fall Below $1/GB By 2018 40% LOWER EACH YEAR Consumer Electronics Keep Driving FLASH FAB Capacity Up. As A Result, FLASH NAND Prices Will Continue To Fall 9
New Technologies Are Emerging Phase-change memory with several advantages over Flash Devices can store and retrieve data 100 times faster Extremely durable, capable of being rewritten 10M times Similar technology already used in rewritable optical storage Greater potential for miniaturization (10nm per cell) Source: Economist, September 1st, 2012. 10
The FLASH 1st Data Management Strategy Hot Data On Fast FLASH SSDs Cold Data On Dense Disks Data Activity Hot High Activity Highly Active Data Is Stored On FLASH SSDs For Fastest Response Time FLASH SSD As Data Ages, Activity Falls, Triggering Automatic Movement To High Capacity Disk Drives For Lowest Cost High Cap. HDD Movement Trigger Cold Low Activity Data Age (5 years) 11
Gaining Economic Efficiency FLASH SSDs For Active Transactions & 3TB HDD For Overflow IOPS per $ 1,2 1 0,8 0,6 0,4 0,2 0 FLASH SSD 15K HDD FAST VP 3TB HDD 0 0,5 1 1,5 2 2,5 GB per $ FLASH SSDs Offers Highest Transaction Efficiency Large Capacity HDD Offers Highest Capacity Efficiency High RPM HDDs Are Not Capacity- Efficient Nor Are They Performance- Efficient FAST VP Automatically And Seamlessly Optimizes Data Across All Disk Technologies 12
Optimizing Your VNX Deployment Lowest $/IOPS And Lowest $/GB Transactions Max System IOPS Add SSDs For More IOPS System Scale Add HDDs For More Capacity 13
The FAST Suite Dynamically Optimizes IO For FLASH 1 st At A 64K Page Size FAST Cache - FAST VP Relationship Controller FAST Cache Caches Data From The HDD Or NL-HDD Tiers In The Pool Operates At A Page Granularity Of 64K FAST VP 64K FAST Cache 64K Dynamically Moves Data Between Tiers In The Storage Pool Operates At A Slice Granularity Of 1GB SSD HDD 64K NL-HDD FAST Virtual Pool 1G Deploying Both Together Ensures Maximum IO Granularity 14
IO Skew Random Slice Activity 1000 900 800 700 600 500 400 300 200 100 0 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 1000 900 800 700 600 500 400 300 200 100 0 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 All 80% Slice Activity 100% Randomized Slice Activity Stack Ranked According To Access Frequency Of A Given Slice 51 Slices Of 101 Has 80% Of IO 15
IO Skew Random Slice Activity Weighted By 3% Cooling Rate 1000 900 800 700 600 500 400 300 200 100 0 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 All "80%" Stack Ranked Slice Map Weighted By A 3% Cooling Rate Now Only 31 Slices Of 101 Total Has 80% Of IO 16
Examples of IO Skew Driven By Data Growth And Business Models Very High Skew Growth: 60% Days Hot: 10 1% Of Data = 80% Of IO Std. Skew Growth: 50% Days Hot: 60 5.5% Of Data = 80% Of IO High Skew Growth: 100% Days Hot: 30 4.2% Of Data = 80% Of IO Low Skew Growth: 50% Days Hot: 90 8.2% Of Data = 80% Of IO 17
How Much FLASH? Used Dynamically With FAST Size of FLASH FIFO Amount Of Data Being Created Daily FLASH Capacity Number Of Days Of High Data Activity Data Follows A Predictable Decay In Activity Older Data Is Constantly Being Replaced By New Highly Active Data The Amount Of FLASH Required Is Determined By: The Amount Of Data Created Each Day, And The Period Of Time It Takes To Cool 18
Modeling Data Decay A Simple GB-Day State Model Describes Typical Data Behavior Probability Of Data Access Hot Where: P = (1-C) A, And A = LOG 1-C (P) P Is Probability Of Data Access C Is Daily Cooling Rate A Is Data Age In Days 1GB Data Age In Days 19
Modeling Data Decay A Simple Gb-day State Model Describes Typical Data Behavior P = (1-C) A, And A = LOG 1-C (P) Hot State Change Where: P Is Probability Of Data Access C Is Daily Cooling Rate A Is Data Age In Days Cold Hot Cold 1GB Data Age In Days 20
Basic Data Decay Model FAST FLASH Tier Cache 20% 60% 15K 1% The Basic Storage Pool Distribution Model Assumes FLASH As The First Line Of Defense For 80% Of All IOs As An Additional Performance Buffer, 15K Drives Are Used For Overflow For 19% Of All IOs Only A 1% Chance Remains That Any IOs Will Be Serviced From Slow 2TB Drives 7 days 52days 250 days 21
How Long Does Data Stay Hot? Depends On Business Model, Applications And Workload Hot Days Daily Cooling Rate @ 80% 30 5.2% 60 2.7% 90 1.8% 120 1.4% Cold Hot 30 days 60 days 90 days 120 days Days Of High Data Activity 22
FLASH Percentage A Function Of Data Growth And Service Level FLASH Portion Data Growth Rate Relative More FLASH Is Needed When: Service Level (FLASH Hit Rate) Is Elected High, And Data Growth Is High More Than 25% FLASH Is Highly Unlikely! 23
Getting The Right Blend Of Flash 3 Fundamental Business Questions How Much Data Is Under Management Today? 30 TBs How Much Is Your Data Growing Each Year? 50% Y/Y How Long Does Your Data Stay Hot? 60 days 24
Calculating Net New Data Net New Is A Function Of Amount Of Starting Data And Growth 50% Growth Rate 15 30 30 50% Of 30TB Is 15TB The Average Amount Of Data Generated Each Day: 15 X 1024 GB / 365 = 42 GB per day Start Year 1 25
Calculating Needed FLASH FLASH Only Needs To Be Large Enough To Hold Hot Data FLASH Portion 2,5 Needed FLASH 12,5 30 30 Start Year 1 FLASH Capacity: 60 Days X 42 GB = 2,520 GB Flash Percentage: FLASH Capacity/Total Capacity 2,520 GB/(45 X 1024) GB X 100 = 5.4% 26
Calculating How Much FLASH FLASH Portion FLASH % = Yearly Growth Rate% X Number of Hot Days X 100 365 X (Yearly Growth Rate% + 100%) FLASH portion of storage pool 20% 15% 10% 5% 0% FLASH Portion As A Function Of Yearly Data Growth 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 30 days Hot 1% 1% 2% 2% 3% 3% 3% 4% 4% 4% 60 days Hot 1% 3% 4% 5% 5% 6% 7% 7% 8% 8% 90 days Hot 2% 4% 6% 7% 8% 9% 10% 11% 12% 12% 120 days Hot 3% 5% 8% 9% 11% 12% 14% 15% 16% 16% 27
Using The Business Data To Configure With No Trace Data, The Model Is Used Hot SSD 60 days 178 days 3-Tier FLASH 1 st Strategy The SSD Tier Retains A 80% Probability Of Data Re-use An 15K HDD Tier Is Used As Overflow With A 19% Chance Of Data Re-use Finally, The NL-HDD Tier Is Used For Long Term Online Storage With A 1% Chance And Lower Of Data Re-use Cold 15K-HDD NL-HDD 28
Configuration Example 30TB Usable, 50% Growth, 60 Days Hot @ 80% FLASH Service VNX5500 Large Foot Print 200 SAS 15K HDDs $380,573 VNX5300 Smaller Foot Print SSDs For FAST Cache 4 SSDs For FAST VP 20 15K SAS HDDs For FAST VP 15 NL-SAS HDDs For FAST VP 24 Total 63 $182,882 Monolithic 29
Tangible Business Benefits Dramatically Lower Cost And Higher Transaction Support $ Per Usable GB Yearly Power Consumption With FAST $3.97 With FAST 7.030 Without FAST $8.26 Without FAST 30.660 Disk Slots Needed Transactions Supported With FAST 63 With FAST 36,686 Without FAST 200 Without FAST 15,050 30
Virtualization Further Pressures Storage Moore s Law And Better CPU Utilization Drives IOPS Need 32 cores 1M IOPS 32 cores 1M IOPS 2M IOPS 32 cores 1M IOPS 32 cores 1M IOPS NEED For SPEED Note: 2M host pressure IOPS is based on 50% storage side IOPS per host 200K IOPS Application Management - Like VMotion Requires Shared Storage A Performance Budget Of 1M IOPS Per Host Is Now A Reality Shared Storage Risks Becoming Bottleneck In High Performance Virtualization Deployments 31
VNX Virtualized Cloud Architecture Host-Side Read Caching Multiplies Performance Performance Scale 2M IOPS 10X BETTER PERFORMANCE Note: 2M storage IOPS is based on 90%/10% read/write mix Capacity Scale Scale Out For Hot Data And Read Performance With FLASH Based Host-Side Caching Scale Up For Deep Archive Of Cold Data All Writes Are Protected By Synchronous Cache Mirroring To DRAM & FLASH In Central Array 32
High Performance Storage Hierarchy The Data Continuum Is FLASH Centric And Extends To The Host PCIe FLASH SSD FLASH 50uS 200uS Working Set Host-Side Read FLASH Caching Shields Data From SAN And Array Latency Array-Side FLASH Protects Data Against Host Failure Array-Side HDDs Accumulates Inactive Data At Lowest Cost HDD 1,500uS Inactive Data 33
The IO Spread: Access Density PCIe FLASH Offers 8,333X Better Data Access Than 15K HDD 10000 1000 100 10 1 0,1 0,01 0.03 2.500 25 0,3 NL-HDD HDD SSD PCIe FLASH Access Density: IOPS/GB 34
Data Continuum As Data Activity Falls, Different Technologies Apply Hot High Activity VFCache Data Activity SSD FAST Cache SSD FAST VP COST OPTIMIZED Dynamic Storage Hierarchy Differentiated On Access Density Cost 15K HDD NL-HDD Cold Low Activity Data Age 35