Increase MySQL* Database Performance with Intel plus Cache Acceleration Software David Tuhy - Senior Director of Business Development Intel Non-Volatile Memory Solutions Group
Performance Moore s Law Provides Exponential Performance Growth CPU 2005 2007 2009 2011 2013 Source: Intel NSG Storage Technologies Group 2
Performance Increasing Gap HDD performance not keeping pace CPU 2005 2007 2009 2011 2013 HDD Source: Intel NSG Storage Technologies Group Storage becomes the performance bottleneck 3
Performance Closing the Gap with Solid State Disks (s) CPU HDD 2005 2007 2009 2011 2013 Source: Intel NSG Storage Technologies Group s dramatically reduce the performance gap 4
MySQL* Server Database Acceleration Sysbench* Read/Write Sysbench is an open source Linux workload generator that runs on MySQL* simulating Database transaction and response times: 20% Caching P3700 Workload Completes in 7s Click here to run S3710 Workload Completes in Video 11s snapshot of the Demo Intel CAS & P3700 Workload Completes in 32s 100% NVMe 100% SATA Intel CAS with 20% NVMe 100% HDD in RAID-5 Configuration HDD Workload Completes in 89s Configuration Common to All Tests: Supermicro SuperServer 2028U-TR4+, dual socket Intel Xeon E5-2699 v3, 128 GB DDR4 RAM, Hardware RAID controller Database: 750GB Sysbench database Operating System: Redhat Enterprise Linux 6.7, Intel x86-64, kernel 2.6.32-573. Configuration 1- HDD Time to Complete Workload: 89s 4ea 15k RPM (300GB Seagate Savvio ST930065355) SAS HDDs in RAID5 (LSI MegaRAID SAS-3 3108 Controller) Configuration 2 - HDDs + Intel CAS & Intel P3700 Time to Complete Workload: 32s 4ea 15k RPM SAS HDDs in RAID5 (LSI MegaRAID SAS-3 3108 Controller) Intel DC P3700 1.6TB with Intel CAS caching (20% of Database = 150GB cache partition) Time in Seconds Technology Demonstration: Intel CAS on an Intel P3700 NVMe as a Caching Device Reduces Workload Execution Time by 64% Configuration 3 Intel S3710 Time to Complete Workload: 11s 1ea Intel DC S3710 1.2TB Configuration 4 Intel P3700 Time to Complete Workload: 7s 1ea Intel DC P3700 1.6TB Data collected at 4/12/2016 Intel technologies may require enabled hardware, specific software, or services activation. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSbench, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. *Other names and brands may be claimed as the property of others 5
Simplified Anatomy of a Solid State Drive () U.2 NVM Technology NVM Technology NVMe or SATA or SAS Controller ASIC Firmware NVM Technology NVM Technology NVM Technology NVM Technology NVM Technology NVM Technology Intel 2015
Non-Volatile Memory (NVM) Technology Transformations 3D NAND MLC and TLC Building block enabling expansion of into HDD segments 3D Xpoint Building blocks for ultra high performance storage & memory
Not all 3D NAND is the same Moving to OTHER 3D NAND 2 INTEL 3D NAND Intel FG cell has a smaller footprint 2D NAND ~1.3X faster Read and Write times Increased power efficiency 3D NAND Improved endurance enabling TLC at multiple endurance levels Superior data integrity RBER, cycling, and retention 1 Breakthrough cost > 25 years of experience Scale faster & bigger >10TB in a Denser memory array More efficient utilization Intel CMOS under array More Gb/mm 2 384Gb 1 st Gen vs. 256Gb 3 rd Gen 1. 3D NAND benefits. Source Intel datasheets. Comparing tprog and Tread for 20nm 2D NAND to 3D NAND gen. 1. http://www2.techinsights.com/l/8892/2014-07-29/frm3m
Latency (µs) 10,000 NVMe is the next quantum leap for s 100 175 150 125 100 NAND technology offers ~100X reduction in latency versus HDD NVMe eliminates ~20 µs of latency today 75 50 25 Drive Latency Controller Latency (ie. SAS HBA) Software Latency 0 HDD +SAS/SATA NAND +SAS/SATA NAND +NVMe Technology claims are based on comparisons of latency, density and write cycling metrics amongst memory technologies recorded on published specifications of in-market memory products against internal Intel specifications.
New NVM Storage Class Memory 3D Xpoint TECHNOLOGY Breaks The Memory Storage Barrier SRAM Latency: 1X Size of Data: 1X DRAM Latency: ~10X Size of Data: ~100X 3D XPoint Latency: ~100X Size of Data: ~1,000X STORAGE NAND Latency: ~100,000X Size of Data: ~1,000X HDD Latency: ~10 MillionX Size of Data: ~10,000X Technology claims are based on comparisons of latency, density and write cycling metrics amongst memory technologies recorded on published specifications of in-market memory products against internal Intel specifications.`
3D Xpoint Outperforms 3D NAND Technology Latency (µs) 90 35 30 25 20 15 Drive Latency Controller Firmware NVM Tread NVM xfer 3D XPoint based s provide ~10x latency reduction vs NAND s Software latency dominates and storage requires large data access granularity DIMM memory eliminates software latency and permits small granularity cache line access 10 5 Access time to <1µs Drive Latency Software Latency 0 NAND +NVMe 3D XPoint +NVMe DIMM Memory Technology claims are based on comparisons of latency, density and write cycling metrics amongst memory technologies recorded on published specifications of in-market memory products against internal Intel specifications.
MySQL* Benchmarking Setup: Hardware Config, Software Config, and Workloads MySQL Intel CAS SAN/NAS DAS *Other names and brands may be claimed as the property of others. 1 Single - Single MySQL database in a Single Server with Direct Attached Storage (DAS) Hardware Config: Comparison between each of the following 4-Disk RAID5 of HDDs (Seagate Savvio 300GB 15k SAS) Single SATA (Intel S3710 1.2TB) Intel CAS (Intel P3700 1.6TB w/150gb cache partition, accelerating 4-Disk RAID5 of HDDs) Single NVMe (Intel P3700 1.6TB) Software Config: RHEL 6.7 + 2.6.32-573 kernel CAS for Linux v3.0 750GB MySQL database on each of the 4 hardware configurations Workloads: Sysbench* OLTP read mode (100% reads) Sysbench* OLTP default mode (~75/25 random read/write) 12
MySQL* Server Database Acceleration Sysbench* Read Sysbench* is an open source Linux workload generator that runs on MySQL* simulating Database transaction and response times: 20% Caching P3700 Workload Completes in 6s S3710 Workload Completes in 8s Intel CAS & P3700 Workload Completes in 19s Configuration Common to All Tests: Supermicro SuperServer 2028U-TR4+, dual socket Intel Xeon E5-2699 v3, 128 GB DDR4 RAM, Hardware RAID controller Database: 750GB Sysbench database Operating System: Redhat Enterprise Linux 6.7, Intel x86-64, kernel 2.6.32-573. Configuration 1- HDD Time to Complete Workload: 89s 4ea 15k RPM (300GB Seagate Savvio ST930065355) SAS HDDs in RAID5 (LSI MegaRAID SAS-3 3108 Controller) HDD Workload Completes in 89s Configuration 2 - HDDs + Intel CAS & Intel P3700 Time to Complete Workload: 19s 4ea 15k RPM SAS HDDs in RAID5 (LSI MegaRAID SAS-3 3108 Controller) Intel DC P3700 1.6TB with Intel CAS caching (20% of Database = 150GB cache partition) Configuration 3 Intel S3710 Time to Complete Workload: 8s 1ea Intel DC S3710 1.2TB Time in Seconds Technology Demonstration: Intel CAS on an Intel P3700 NVMe as a Caching Device Reduces Workload Execution Time by 79% Configuration 4 Intel P3700 Time to Complete Workload: 6s 1ea Intel DC P3700 1.6TB Data collected at 4/12/2016 Intel technologies may require enabled hardware, specific software, or services activation. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSbench, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. *Other names and brands may be claimed as the property of others 13
Read intensive MySQL* results Benefit of Intel CAS with Intel NVMe X X X Baseline X Intel CAS Read Results ~4.7x Transactions per Second (TPS) performance improvement over 4x HDD in a RAID 5 (parity) configuration 79% reduction in total workload runtime Best performance is all NVMe See configuration for testing on page 12 and Page 13 14
MySQL* Server Database Acceleration Sysbench* Read/Write Sysbench is an open source Linux workload generator that runs on MySQL* simulating Database transaction and response times: 20% Caching NVMe P3700 Workload Completes in 7s SATA S3710 Workload Completes in 11s Intel CAS & P3700 Workload Completes in 32s HDD Workload Completes in 89s Configuration Common to All Tests: Supermicro SuperServer 2028U-TR4+, dual socket Intel Xeon E5-2699 v3, 128 GB DDR4 RAM, Hardware RAID controller Database: 750GB Sysbench database Operating System: Redhat Enterprise Linux 6.7, Intel x86-64, kernel 2.6.32-573. Configuration 1- HDD Time to Complete Workload: 89s 4ea 15k RPM (300GB Seagate Savvio ST930065355) SAS HDDs in RAID5 (LSI MegaRAID SAS-3 3108 Controller) Configuration 2 - HDDs + Intel CAS & Intel P3700 Time to Complete Workload: 32s 4ea 15k RPM SAS HDDs in RAID5 (LSI MegaRAID SAS-3 3108 Controller) Intel DC P3700 1.6TB with Intel CAS caching (20% of Database = 150GB cache partition) Configuration 3 Intel S3710 Time to Complete Workload: 11s 1ea Intel DC S3710 1.2TB Time in Seconds Technology Demonstration: Intel CAS on an Intel P3700 NVMe as a Caching Device Reduces Workload Execution Time by 64% Configuration 4 Intel P3700 Time to Complete Workload: 7s 1ea Intel DC P3700 1.6TB Data collected at 4/12/2016 Intel technologies may require enabled hardware, specific software, or services activation. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSbench, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. *Other names and brands may be claimed as the property of others 15
Read/Write MySQL results Benefit of Intel CAS with an Intel NVMe X X X Baseline Intel CAS Read/Write Results ~2.8x TPS performance improvement over 4x HDD in a RAID 5 (parity) config. 64% reduction in total workload runtime Best Performance is all NVMe X See configuration for testing on page 12 and Page 15 16
MySQL * Supported System Topologies for Intel Cache Acceleration Software (CAS) 1 2 Single - Single MySQL database in a Single Server with Direct Attached Storage (DAS) SAN/NA Intel S MySQL DAS CAS Multiple s, Single MySQL database with Intel CAS and, external Shared Storage MySQL Intel CAS External Shared Storage node *Other names and brands may be claimed as the property of others. Multiple s Source SuperMicro* Server Graphics 17
MySQL* Supported Models for Intel Cache Acceleration Software (CAS) 3 4 Single Single MySQL database in a single server with external storage using CAS+ MySQL Intel CAS External Shared Storage node Multiple s, Multiple MySQL, with Intel CAS+ in external storage MySQL MySQL MySQL Intel CAS External shared storage node *Other names and brands may be claimed as the property of others. Source SuperMicro* Server Graphic 6
Do not do this configuration with Intel Cache Acceleration Software version 3.0 Intel CAS Intel CAS MySQL External Shared Storage Issue: CAS cannot see updates made to the central storage of another CAS instance. Intel CAS MySQL Intel CAS Storage node Shared Storage Need to avoid data integrity issues. MySQL MySQL Intel CAS Intel CAS Server 7
For more information on Intel Cache Acceleration Software Intel Cache Acceleration Software website http://www.intel.com/content/www/us/en/software/intel-cache-acceleration-softwareperformance.html Intel Cache Acceleration Software 120 day free trial https://www-ssl.intel.com/content/www/us/en/forms/cache-acceleration-contact-us.html 20