Comments
Transcript
Why POWER8 is the platform of choice for Linux
Why POWER8 is the platform of choice for Linux Gary Andrews [email protected] 03/26/2015 IBM Competitive Project Office Proving the Value of IBM Technology © 2015 IBM Corporation Competitive Project Office POWER8 and Linux Provide the best customer value Purpose Built Best Value 1. Architecture Matters Better Performance 2. Best Price / Performance 1. 2. 3. 4. 5. 6. 2 Java Applications Workload density Lower Cost Mixed workloads OLTP Analytics Better Service Big Data © 2015 IBM Corporation Competitive Project Office Power Architecture is Purpose Built POWER8 design point is for big workloads: Analytics Java Big Data POWER8 OLTP ERP Cloud Social Mobile Intel design point is for multiple markets: Smart Phones, Ultra Books, Desktop, Servers 3 © 2015 IBM Corporation Competitive Project Office POWER8 – Significant Improved Technology Leadership that is Ready for Bigger Workloads A purposeful balanced design that delivers new record performance More Cores More Threads 12 processor cores per socket (50% more than before) that deliver 2X better per core performance SMT8 – 8 dynamic threads per core, supporting SMT1, 2, 4, & 8 modes dynamically across VMs What this means Better scale up performance, and more throughput per scale out server node What this means You choose – Deploy VM’s in the optimal SMT mode based on application needs More Cache More Bandwidth 3X the on-chip cache as POWER7 – plus 128MB of new off-chip cache as well 3.5X more memory and 2.8X more I/O bandwidth than POWER7 What this means Memory-intensive applications will perform better as memory latency is reduced What this means Data-hungry applications (like Big Data & Analytics) will respond twice as fast and scale more efficiently. 4 © 2015 IBM Corporation Competitive Project Office Architecture Matters When You Design A Micro Processor For Emerging Big Workloads It’s not about the number of transistors, it is what you do with them to handle Big Workloads POWER8 vs. Ivy Bridge EX POWER7 to POWER8 1.2 Billion Transistors 45nm to 22nm 567 mm 4.2 Billion Transistors 650 mm 2 Westmere EX to Ivy Bridge EX 2.6 Billion Transistors 513 mm 5 32nm to 22nm 4.3 Billion Transistors 541 mm 2 - 96 threads/socket vs. 30 - 4x Memory Bandwidth - 3x on-die Cache - Cache latency reduced by 50% - 5x I/O Bandwidth - 15 metal layers vs. 9 - eDRAM vs SRAM POWER8 Unique Technology - CAPI Technology - L4 Cache - Dynamic Overclocking © 2015 IBM Corporation Competitive Project Office POWER8 Is Designed for Superior Performance Sandy Bridge EP Ivy Bridge EP Ivy Bridge EX Haswell EP E5-x6xx E5-26xx v2 E7-88xx v2 E5-26xx v3 1.7-3.7 1.9-3.4 1.6-3.5 3.0-4.35 GHz 1,2* 1, 2* 1, 2* 1, 2* 1, 2, 4, 8 Max Threads / sock 16 24 30 36 96 Max L1 Cache 32KB 32KB* 32KB* 64 KB 64KB Max L2 Cache 256 KB 256 KB 256 KB 256KB 512 KB Max L3 Cache 20 MB 30 MB 37.5 MB 45 MB 96 MB Max L4 Cache 0 0 0 0 128 MB 31.4-51.2 GB/s 42.6-59.7 GB/s 68-85** GB/s 51-68 GB/s 230 - 410 GB/sec Clock rates (GHz) SMT options Memory Bandwidth 1.8–3.6 POWER8 * Intel calls this Hyper-Threading Technology (No HT and with HT) *32KB running in “Non-RAS mode”, Only 16KB with ECC Error correction **85GB running in “Non-RAS mode” = dual-device error NOT supported 6 © 2015 IBM Corporation Competitive Project Office Intel’s Performance per Core is Not Increasing Over Previous Generation RPE2** per Core 2500 2 Socket HP Servers 2283 2069 2049 1921 2000 1500 1000 Sandy Bridge EP 2.9 GHz 16 cores Ivy Bridge EP 2.7 GHz Ivy Bridge EX 2.8 GHz Haswell EP 2.3 GHz 24 cores 30 cores 36 cores 500 0 The number shown is best in each category (sockets and number of cores) **Gartner RPE2 Details: http://www.gartner.com/technology/research/RPE2-methodology-details.jsp RPE2** numbers are derived from the following six benchmark inputs: SAP SD Two-Tier, TPC-C, TPC-H, SPECjbb2006 and two SPEC CPU2006 components The data on this chart is derived from RPE2 from Gartner, Inc.'s Competitive Profile tool. © 2014 Gartner, Inc. and/or its affiliates. All rights reserved. 7 © 2015 IBM Corporation Competitive Project Office POWER8 per Core Performance is Increasing! 3500 This was a POWER8 Design Goal 1500 1000 500 POWER7+ 2000 POWER8 2500 POWER7 RPE2** per core 3000 0 730 3.55 GHz 16 cores 2 sockets 730+ 4.20 GHz 16 cores 2 sockets The number shown is best in each category (sockets and number of cores) **Gartner RPE2 Details: http://www.gartner.com/technology/research/RPE2-methodology-details.jsp S824 3.52 GHz 24 cores 2 sockets RPE2** numbers are derived from the following six benchmark inputs: SAP SD Two-Tier, TPC-C, TPC-H, SPECjbb2006 and two SPEC CPU2006 components The data on this chart is derived from RPE2 from Gartner, Inc.'s Competitive Profile tool. © 2014 Gartner, Inc. and/or its affiliates. All rights reserved. 8 © 2015 IBM Corporation Competitive Project Office Why Is This Important? POWER8 Design Focus and Results - More threads, cache - More memory bandwidth - More I/O bandwidth - Fastest performance for all workloads - Ready to address new Big Data and Analytic workloads Higher Performance per core - Lower Software costs and TCA - Fewer servers, lower support costs CAPI - New Accelerator technologies Intel Design Focus and Results - More sockets per server More cores per server Performance per core remains the same or is less 9 - Faster overall SERVER performance - Overall solution cost is higher © 2015 IBM Corporation Competitive Project Office 10 © 2015 IBM Corporation Competitive Project Office POWER8 and Linux Provide the best customer value Purpose Built Best Value 1. Architecture Matters Better Performance 2. Best Price / Performance 1. 2. 3. 4. 5. 6. 11 Java Applications Workload density Lower Cost Mixed workloads OLTP Analytics Better Service Big Data © 2015 IBM Corporation Competitive Project Office Client Tier These user interactions are repeated by thousands of simulated users RPT Workbench Deploy 1. Friendly Bank home page 2. Logon 3. Check account balances 4. Transfer funds (2) 5. View transaction history 6. View profile 7. Logoff Stats NETWORK Test Harness Setup: POWER8 vs. Intel Ivy Bridge Using Friendly Bank WAS Application RPT Agent RPT Agent Drive Application RPT Agent RPT Agent RPT Agent RPT Agent RPT Agent Database Tier Server Tier System Under Test PowerVM Friendly VMware ESXi WebSphere WebSphere Bank Friendly WebSphere Application Bank Friendly WebSphere Application Bank Friendly WebSphere Application Bank Friendly WebSphere Application Bank Friendly WebSphere Application Bank Friendly Application Bank Application Virtual Machines 12 DB2 DB2 DB2 DB2 DB2 DB2 DB2 DB2 © 2015 IBM Corporation Competitive Project Office Throughput per Core POWER8 SMT delivers more performance per core than Intel Ivy Bridge SMT8 SMT4 SMT2 2.1 – 2.3x SMT1 Intel Ivy Bridge EP (Hyper-Threading – Maximum) Time in Seconds This is an IBM internal study designed to replicate a typical IBM customer workload usage in the marketplace. It consists of a POWER8 S824 with 24 cores, 3.52 GHz, 256GB Memory, AIX 7.1 TL3 SP3, WAS 8.5.5.1, DB2 9.7, JDK 7.0 FP1 compared to an Ivy Bridge EP 24 cores 2.7 GHz, 256 GB Memory, RHEL 6.5, WAS 8.5.5.1, DB2 9.7, JDK 7.0 FP1. The results were obtained under laboratory conditions, and not in an actual customer environment. IBM's internal workload studies are not benchmark applications, nor are they based on any benchmark standard. As such, customer applications, differences in the stack deployed, and other systems variations or testing conditions may produce different results and may vary based on actual configuration, applications, specific queries and other variables in a production environment. 13 © 2015 IBM Corporation Competitive Project Office POWER8 and Linux demonstrate SMT throughput improvement compared to Ivy Bridge Throughput on 24-core servers POWER8 with RHEL SMT = Simultaneous Multi-Threading 200000 2.1X 180000 188,184 171,920 160000 140000 131,696 120000 100000 90,760 80000 60000 40000 20000 0 SMT 1 SMT 2 SMT 4 SMT 8 This is an IBM internal study designed to replicate a typical IBM customer workload usage in the marketplace. It consists of a POWER8 S824 with 24 cores, 256GB Memory, 3.52 GHz, RHEL 7.0, WAS 8.5.5.2, DB2 9.7, JDK 7.0 FP1 compared to an Ivy Bridge EP 24 cores, 256GB Memory, 2.7 GHz, RHEL 6.5, WAS 8.5.5.1, DB2 9.7, JDK 7.0 FP1. The results were obtained under laboratory conditions, and not in an actual customer environment. IBM's internal workload studies are not benchmark applications, nor are they based on any benchmark standard. As such, customer applications, differences in the stack deployed, and other systems variations or testing conditions may produce different results and may vary based on actual configuration, applications, specific queries and other variables in a production environment. Prices, where applicable, are based on published US list prices for both IBM and competitor, and the Total Cost of Acquisition (TCA) includes the list HW and SW prices and 3 years of service & support which is then divided by the number of transactions to get $ per user interaction per second. 14 © 2015 IBM Corporation Competitive Project Office Linux on POWER8 with WAS Delivers Over TWICE the Throughput Compared to Ivy Bridge at 49% Lower Cost Web Application Power S824 Linux WAS WAS ….. RHEL RHEL WAS WAS 4 VMs 188,184 User Interactions per second 2.2x Faster RHEL RHEL $3.15 PowerVM DB2 2S/24 Core POWER8 (3.52 GHz) per UI per sec WebSphere on platform Database off platform 49% Lower cost per UI per sec Online Banking Workload v3.6 15 © 2015 IBM Corporation Competitive Project Office POWER8 and Linux Provide the best customer value Purpose Built Best Value 1. Architecture Matters Better Performance 2. Best Price / Performance 1. 2. 3. 4. 5. 6. 16 Java Applications Workload density Lower Cost Mixed workloads OLTP Analytics Better Service Big Data © 2015 IBM Corporation Competitive Project Office How many virtual machines can you run on a single server? It depends…… 3. Variance of the Workload 1. Size of the Workload High Variability Performance Interactions per second Large Small Workload 2. Size of the Server 16 cores 24 cores 48 cores Low Variability Workload Density … Frequency of Occurrence 4. Workload Management Mixed Workloads Hypervisor Low Priority High Priority Number of cores 17 © 2015 IBM Corporation Competitive Project Office POWER8 Packs Up To 2.3x More Virtual Machines Than Intel On Same Number Of Cores Online Banking Workload Ivy Bridge-EP (24-cores 2.7 GHz) & Competitive Hypervisor 90 82 Number of VMs per server 80 POWER8 (24-cores 3.52 GHz) & PowerVM PowerVM / RHEL vs. Competitive Hypervisor / RHEL 70 60 50 2.3x 40 30 36 23 20 2.1x 10 11 0 728 1960 10 6 Workload Size – User Interactions per second 1.6x 3920 This is an IBM internal study designed to replicate a typical IBM customer workload usage in the marketplace. It consists of a POWER8 S824 with 24 cores, 3.52 GHz, RHEL 6.5, WAS 8.5.5.1, DB2 9.7, JDK 7.0 FP1. compared to an Ivy Bridge EP 24 cores 2.7 GHz, RHEL 6.5, WAS 8.5.5.1, DB2 9.7, JDK 7.0 FP1 and a Competitive hypervisor. The results were obtained under laboratory conditions, and not in an actual customer environment. IBM's internal workload studies are not benchmark applications, nor are they based on any benchmark standard. As such, customer applications, differences in the stack deployed, and other systems variations or testing conditions may produce different results and may vary based on actual configuration, applications, specific queries and other variables in a production environment. 18 © 2015 IBM Corporation Competitive Project Office POWER8 with PowerVM Supports 2.3x More Large Workloads Than Ivy Bridge with Competitive Hypervisor Web Application Power S824 Linux WAS WAS ….. RHEL RHEL WAS WAS 82 VMs 82 Workloads 2.3x $7,218 per Workload Workloads RHEL RHEL PowerVM DB2 50% 2S/24 Core POWER8 (3.52 GHz) Online Banking Workloads each running WebSphere on platform, Database off platform for both tests 19 More 728 User Interactions/Sec Lower cost per Workload © 2015 IBM Corporation Competitive Project Office POWER8 and Linux Provide the best customer value Purpose Built Best Value 1. Architecture Matters Better Performance 2. Best Price / Performance 1. 2. 3. 4. 5. 6. 20 Java Applications Workload density Lower Cost Mixed workloads OLTP Analytics Better Service Big Data © 2015 IBM Corporation Competitive Project Office Workload Prioritization Test Configuration – Apply a Constant and Variable Workload POWER8 E870 Low Priorities 0.25 Entitlement 5 VPs High Priorities 4.75 Entitlement 5 VPs WAS DB2 . LPARs ...... WAS DB2 RHEL 7 RHEL 7 PowerVM PowerVM 40 Cores POWER E870 - 4.19 GHz Intel Ivy Bridge EX High Priorities 950,000 Shares 10 vCPUs WAS DB2 Low Priorities 50,000 Shares 10 vCPUs . .VMs ..... WAS DB2 RHEL 7 RHEL 7 VMWare ESXi 5.5 60 Cores Intel Ivy Bridge EX - 2.80 GHz This is an IBM internal study of a on-line banking workload in a controlled laboratory environment. The IBM system consists of one POWER8 E870 with 40 cores @ 4.19 GHz and 1 TB memory running RHEL 7.0, WAS ND 8.5.5.3, JDK 7.1 and DB2 v10.5 FP3. The Intel-based system consists of one Ivy Bridge EX (E7-4890 v2) with 60 cores @ 2.8 GHz and 1 TB memory running RHEL 7.0, WAS ND 8.5.5.3, JDK 7.1, and DB2 v10.5 FP3. Customer applications, differences in the systems deployed and other system variations or testing conditions may produce different results. 21 © 2015 IBM Corporation Competitive Project Office High Priority Web Workload Running Standalone – Constant Load Constant Load E870 40 cores High Priority throughput: 98.16 Million transactions / 60 min 2.1x more POWER8 transactions vs. Intel 22 © 2015 IBM Corporation Competitive Project Office High Priority Workload on E870 has Minimal Degradation when Low Priority Workload Added E870 40 cores Constant Load High Priority Workload - 6.0% throughput reduction due to workload mixing - 5.0% given up for low priority workload entitlements 98.16M 87.36M (High Priority) 2.04M (Low Priority) 5.8x More Efficient This is an IBM internal study of a on-line banking workload in a controlled laboratory environment. The IBM system consists of one POWER8 E870 with 40 cores @ 4.19 GHz and 1 TB memory running ARHEL 7.0, WAS ND 8.5.5.3, JDK 7.1 and DB2 v10.5 FP3. The Intel-based system consists of one Ivy Bridge EX (E7-4890 v2) with 60 cores @ 2.8 GHz and 1 TB memory running RHEL 7.0, WAS ND 8.5.5.3, JDK 7.1, and DB2 v10.5 FP3. Customer applications, differences in the systems deployed and other system variations or testing conditions may produce different results. 23 © 2015 IBM Corporation Competitive Project Office Run Mixed Workloads on the Same Server to Simplify and Save Money – Variable Load High Priority Workloads Comparison to determine which platform provides the lowest TCA over 3 years with variable loads 34.77M Txn Low Priority Workload High Priority workloads 62.78M Txn/hr 28.01M Txn 17.78M Txn Low priority workloads POWER8 E870 with RHEL 7.0 17.78M Txn/hr IBM WebSphere ND 8.5 IBM DB2 10 ESE High priority online banking workloads driving a total of 62.78 M transactions per hour and low priority discretionary workloads 62.78M Txn High Priority LPARs 17.78M Txn Low Priority LPARs RHEL on Power E870 40 Cores 3 Year TCA: $3.49M 73% lower cost This is an IBM internal study of a on-line banking workload in a controlled laboratory environment. The IBM system consists of one POWER8 E870 with 40 cores @ 4.19 GHz and 1 TB memory running RHEL 7.0, WAS ND 8.5.5.3, JDK 7.1 and DB2 v10.5 FP3. The Intel-based system consists of one Ivy Bridge EX (E7-4890 v2) with 60 cores @ 2.8 GHz) and 1 TB memory running RHEL 7.0, WAS ND 8.5.5.3, JDK 7.1, and DB2 v10.5 FP3. Customer applications, differences in the systems deployed and other system variations or testing conditions may produce different results. Cost analysis based on 3 year total cost of acquisition of hardware, software and support services over a 3 year period. Prices for both IBM and competitor systems based on US list prices valid as of December 2014. 24 © 2015 IBM Corporation Competitive Project Office POWER8 and Linux Provide the best customer value Purpose Built Best Value 1. Architecture Matters Better Performance 2. Best Price / Performance 1. 2. 3. 4. 5. 6. 25 Java Applications Workload density Lower Cost Mixed workloads OLTP Analytics Better Service Big Data © 2015 IBM Corporation Competitive Project Office High level overview of OLTP brokerage workload used in this test Meant to be a more realistic OLTP workload, models a financial/stock brokerage system – Sub-second response times – Multiple access streams Data Application Architecture – – – – Multiple tables, indexes, and data types Requires referential integrity Complex transactions Multiple interacting subsystems Two Metrics can be used 1. Transactions per second (tps) measures total transactions per sec 2. Completed Trades per second measures fully completed transactions 26 © 2015 IBM Corporation Competitive Project Office Competitor DB on Power S824 is 2.1x Faster than PreIntegrated Database Competitor at 44% Lower Cost 600GB Brokerage Workload IBM Power S824 with Database Competitor Power S824 with 24 cores 512 GB RAM AIX 7.1, 64-bit FlashSystem 840 12c 988 $3,326 Completed Trades per second per Completed Trade/sec Competitor DB 2.1x Faster 44% Lower cost per Completed Trade/sec 27 © 2015 IBM Corporation Competitive Project Office POWER8 and Linux Provide the best customer value Purpose Built Best Value 1. Architecture Matters Better Performance 2. Best Price / Performance 1. 2. 3. 4. 5. 6. 28 Java Applications Workload density Lower Cost Mixed workloads OLTP Analytics Better Service Big Data © 2015 IBM Corporation Competitive Project Office The business intelligence concurrent throughput test used in this study 3 Users doing complex reports 1 User 1 User 1 User CR8 CR9 CR13 1 User 2 Users 2 Users 2 Users 3 Users 1 User 2 Users IR1 IR2 IR3 IR4 IR6 IR7 IR12 3 Connections Each user executes its own version of a given report in a fixed duration of 2 hours 29 42 Users doing simple reports 15 Users doing intermediate Reports 15 Connections Data Server 2 Users 12 Users 8 Users IR14 SR5 10 Users 12 Users SR15 SR16 SR10 42 Connections Each report executes one or more SQL queries (or statements) © 2015 IBM Corporation Competitive Project Office Examples of Simple and Complex Queries Example of a typical simple query: 2-3 parameters compared - QUERY 2A Simple_GoBusinessView_Dashboard select distinct "Product_forecast"."YEAR" "Year1" from "GOSL"."PRODUCT_FORECAST" "Product_forecast" Example of a typical complex query: 100-150 parameters compared -- QUERY 3A Complex_TPA_PDF with "Order_method_dimension14" as ( select "Order_method_dimension"."ORDER_METHOD_KEY" "ORDER_METHOD_KEY" , min("Order_method_dimension"."ORDER_METHOD_EN") "Order_method" from "GOSLDW"."ORDER_METHOD_DIMENSION" "Order_method_dimension" group by "Order_method_dimension"."ORDER_METHOD_KEY"), "Product_line15" as ( select "Product_line"."PRODUCT_LINE_CODE" "PRODUCT_LINE_CODE" , min("Product_line"."PRODUCT_LINE_EN") "Product_line" from "GOSLDW"."PRODUCT_LINE" "Product_line" group by "Product_line"."PRODUCT_LINE_CODE"), "Sales_territory_dimension12" as ( select "Sales_territory_dimension"."COUNTRY_KEY" "COUNTRY_KEY" , "Sales_territory_dimension"."COUNTRY_CODE" "COUNTRY_CODE" , "Sales_territory_dimension"."SALES_TERRITORY_KEY" "SALES_TERRITORY_KEY" , "Sales_territory_dimension"."SALES_TERRITORY_CODE" "SALES_TERRITORY_CODE" , "Sales_territory_dimension"."COUNTRY_EN" "COUNTRY_EN" , "Sales_territory_dimension"."FLAG_IMAGE" "FLAG_IMAGE31" , "Sales_territory_dimension"."SALES_TERRITORY_EN" "SALES_TERRITORY_EN" from "GOSLDW"."SALES_TERRITORY_DIMENSION" "Sales_territory_dimension"), "Gender_lookup13" as ( select "Gender_lookup"."GENDER_CODE" "GENDER_CODE" , min("Gender_lookup"."GENDER") "GENDER" from "GOSLDW"."GENDER_LOOKUP" "Gender_lookup" where "Gender_lookup"."LANGUAGE" = 'EN' group by "Gender_lookup"."GENDER_CODE"), "Retailer__model_" as ( select "Retailer_dimension11"."RETAILER_SITE_KEY" "Retailer_site_key" , "Sales_territory_dimension12"."SALES_TERRITORY_KEY" "Sales_territory_key" , "Sales_territory_dimension12"."SALES_TERRITORY_EN" "Sales_territory" from "GOSLDW"."RETAILER_DIMENSION" "Retailer_dimension11", "Sales_territory_dimension12", "Gender_lookup13" where "Retailer_dimension11"."GENDER_CODE" = "Gender_lookup13"."GENDER_CODE" and "Retailer_dimension11"."COUNTRY_KEY" = "Sales_territory_dimension12"."COUNTRY_KEY") select "Order_method_dimension14"."ORDER_METHOD_KEY" "Order_method0key" , "Order_method_dimension14"."Order_method" "Order_method1" , "Product_line15"."PRODUCT_LINE_CODE" "Product_linekey" , "Product_line15"."Product_line" "Product_line0" , "Retailer__model_"."Sales_territory_key" "Retailer_territorykey" , "Retailer__model_"."Sales_territory" "Sales_territory" , cast("Time_dimension17"."CURRENT_YEAR" as char(4)) "Yearkey" , cast("Time_dimension17"."QUARTER_KEY" as char(6)) "Quarterkey" , cast("Time_dimension17"."MONTH_KEY" as char(6)) "Monthkey" , sum("Sales_fact18"."GROSS_PROFIT") "Gross_profit“ from "Order_method_dimension14", "Product_line15", "Retailer__model_", "GOSLDW"."TIME_DIMENSION" "Time_dimension17", "GOSLDW"."SALES_FACT" "Sales_fact18", "GOSLDW"."PRODUCT_TYPE" "Product_type19", "GOSLDW"."PRODUCT_DIMENSION" "Product_dimension20" where "Order_method_dimension14"."ORDER_METHOD_KEY" = "Sales_fact18"."ORDER_METHOD_KEY" and "Product_dimension20"."PRODUCT_KEY" = "Sales_fact18"."PRODUCT_KEY" and "Product_type19"."PRODUCT_TYPE_CODE" = "Product_dimension20"."PRODUCT_TYPE_CODE" and "Product_line15"."PRODUCT_LINE_CODE" = "Product_type19"."PRODUCT_LINE_CODE" and "Time_dimension17"."DAY_KEY" = "Sales_fact18"."ORDER_DAY_KEY" and "Retailer__model_"."Retailer_site_key" = "Sales_fact18"."RETAILER_SITE_KEY" group by "Order_method_dimension14"."ORDER_METHOD_KEY", "Order_method_dimension14"."Order_method", "Product_line15"."PRODUCT_LINE_CODE", "Product_line15"."Product_line", "Retailer__model_"."Sales_territory_key", "Retailer__model_"."Sales_territory", cast("Time_dimension17"."CURRENT_YEAR" as char(4)), cast("Time_dimension17"."QUARTER_KEY" as char(6)), cast("Time_dimension17"."MONTH_KEY" as char(6)) 30 © 2015 IBM Corporation Competitive Project Office DB2 BLU & Cognos on POWER8 delivers analytics reports faster than Ivy Bridge 18x BLU Lightening Test (Telecom Analytics workload) Faster DB2 BLU & Cognos on POWER8 S822L, 3.4 GHz 16-cores active S824, 3.52 GHz 24-cores 31 System Cost $263,730 43,944 Simple Reports per Hour 7,758 Intermediate Reports per Hour Simple Reports 40x Faster Intermediate Reports 212 Complex Reports per Hour 2,267 Simple Reports per Hour 185 Intermediate Reports per Hour 3.0x 0.27 Complex Reports per Hour Lower 747x Faster Complex Reports Price © 2015 IBM Corporation Competitive Project Office POWER8 and Linux Provide the best customer value Purpose Built Best Value 1. Architecture Matters Better Performance 2. Best Price / Performance 1. 2. 3. 4. 5. 6. 32 Java Applications Workload density Lower Cost Mixed workloads OLTP Analytics Better Service Big Data © 2015 IBM Corporation Competitive Project Office BigInsights on POWER8 beats the competition with TeraSort benchmark Normalized Performance GB / Core / Min 1.4 1.7X 1.2 1.27 GB/core/min 1 0.8 0.67 0.6 HP Ivy Bridge EP with Cloudera 0.4 0.73 0.75 SGI Ivy Bridge EP with Cloudera Cisco 0.2 0 Xeon E5-2697v2 8 nodes 192 cores 1 TB Xeon E5-2630v2 24 nodes 288 cores 10 TB Sandy Bridge EP with Apache Hadoop Xeon E5-2665 16 nodes 256 cores 10 TB POWER8 with BigInsights S822L 8 nodes 192 cores 10 TB Cisco Paper - http://www.cisco.com/c/dam/en/us/solutions/collateral/borderless-networks/advanced-services/le_tera.pdf SGI Paper - http://www.sgi.com/pdfs/4440.pdf HP Ivy Bridge with Cloudera and POWER8 with BigInsights were tested in IBM laboratories 33 © 2015 IBM Corporation Competitive Project Office IBM BigInsights with Big SQL delivers Big Data queries faster than Ivy Bridge Modern Business Intelligent workload queries BigInsights v3.0 on 8 S822L data nodes S822L, 3.3 GHz 24 cores 256 GB Memory RHEL 6.5 34 11.2x Time to complete queries in 7 concurrent streams 8 hours 40 min Faster 4.4x Lower Price Performance © 2015 IBM Corporation Competitive Project Office POWER8 and Linux – The Best Value Purpose Built Best Value 1. Architecture Matters Better Performance 2. Best Price / Performance 1. 2. 3. 4. 5. 6. 35 Java Applications Workload density Lower Cost Mixed workloads OLTP Analytics Better Service Big Data © 2015 IBM Corporation