1

2 Performance Innovations with Oracle Database 12cKevin Jernigan Senior Director Product Management System Technologies, Oracle

3 The following is intended to outline our general product directionThe 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.

4 Agenda High Performance TodayIn-Memory & Parallelism for High Performance High Performance with Large Data Volumes End-to-end Performance Architecture

5 High Performance TodayDatabase Scaling

6 SMP Scale-Up Very mature20 years of experience Many customers with largest SMPs on the market 64 to 256 CPUs Up to 2 TB DRAM Exadata, Sun M9000, HP Superdome, IBM Regatta Single System Image Easy to manage Easy to design applications Works great, but eventually hits a wall Need at least two for availability

7 RAC Scale-Out Runs all Oracle database applicationsHighly available and scalable No Idle Resources Single System Image Thousands of production customers

8 Leader in Industry BenchmarksWorld Record Leadership TPC-C Performance Clustered & Non-Clustered Oracle TPC-C Price/Performance 3,000GB Non-Clustered 10,000GB Non-Clustered 30,000 GB SAP Sales and Distribution Parallel SAP Business Intelligence (BI-D) Data Mart As of April 3, Source: SAP, The SAP SD-Parallel standard application benchmark performed on September 5, 2011, by Oracle in Burlington, MA, has been certified with the following data: 180,000 SAP SD-Parallel benchmark users (SAP SD Parallel world record result) 0.63 seconds average dialog response time, 20,327,670 fully processed order line items per hour, 60,983,000 dialog steps per hour, 1,016,380 SAPS, seconds/0.055 seconds average database request time (dialog/update). CPU utilization of servers, 89% (node 1 active: 85%, node 2 active: 89%, node 3 active: 90%, node 4 active: 90%, node 5 active: 90%, node 6 active: 89%, node 7 active: 89%, node 8 active: 90%). Server configuration: 8 Sun Fire X4800 servers each with 8 processors / 80 cores / 160 threads, Intel Xeon Processor E7-8870, 2.40 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 30 MB L3 cache per processor, 512 GB main memory, Solaris 10 operating system, Oracle Database 11g Release 2 with Oracle Real Application Clusters, and SAP enhancement package 4 for SAP ERP 6.0. Certification Number: The SAP Business Intelligence-Data Mart (BI-D) standard application benchmark performed on May, 6, 2010 by Fujitsu in Walldorf, Germany has been certified with the following data: Throughput/hour (query navigation steps): 1,624,629. CPU utilization of central server: 94% (2 nodes active: 93% / 94%). Operating system, central server: SuSE Linux Enterprise Server 10. RDBMS: Oracle Database 11g. Technology platform release: SAP NetWeaver® 7.0 technology platform (non-Unicode). Configuration: 2 server (2 active nodes): Fujitsu PRIMERGY RX600 S5, 4 processors / 32 cores / 64 threads, Intel Xeon Processor X7560, 2.26 GHz, 6.40GT/s Intel® QPI, 24 MB L3 cache per processor, 128 GB main memory and Fujitsu PRIMERGY RX600 S5, 4 processors / 32 cores / 64 threads, Intel Xeon Processor X7560, 2.26 GHz, 6.40GT/s Intel® QPI, 24 MB L3 cache per processor, 256 GB main memory. Certification number: As of 4/4/13: Source: Transaction Processing Performance Council, Oracle Sun SPARC T3-4 server cluster, 30,249,688 tpmC, $1.01/tpmC, available 6/1/11. : SPARC T5-8 (8/128/1024) 8,552,523 tpmC, $0.55 $/tpmC, avail 9/25/13. HP ProLiant ML350 G6, 290,040 tpmC, $.39/tpmC, 4.22 watts/KtpmC, available 8/16/10 (world record TPC-C price/performance). Sun SPARC Enterprise M9000 server, 386,478 9/22/11 (world record TPC-H 3TB non-clustered). HP Integrity Superdome server, 208, available 9/10/08 (world record TPC-H 10TB non-clustered). HP Integrity Superdome Server, 150,960 available 6/18/07 (world's first TPC-H 30TB result, world record TPC-H 30TB. Please see notes page for SAP benchmark information (www.sap.com/benchmark).

9 Best OLTP Performance Source: Transaction Processing Performance Council, as of 9/28/2012. Oracle Sun SPARC T3-4 server cluster, 30,249,688 tpmC, $1.01/tpmC, available 6/1/11. IBM Power 780 Server, 10,366,254 tpmC, $1.38/tpmC, available 10/13/10. HP Integrity Superdome-Itanium2, 4,092,799 tpmC, $2.93/tpmC, available 8/6/07. November 2003: HP Superdome Server, 1,008,144 tpmC, $8.33/tpmC, available 4/14/04 (first TPC-C result to hit One Million., non-clustered result. December 2003: 16-Node HP Integrity rx5670, clustered, 1,184,893 tpmC, $5.52/tpmC, available 4/30/04. (first TPC-C rclustered result to top One Million tpmC).

10 Best OLTP Price-PerformanceBest Single Processor Result Value Leadership Over Microsoft As of September 28, 2012: 10/15/10Source: Transaction Processing Performance Council (TPC), ProLiant ML350 G6, 1 processor, 6 cores, 290,040 tpmC, $.39/tpmC, 4.22 watts/KtpmC, Oracle Database 11g Standard Edition One with OEL, available 8/16/10 (world record TPC-C price/performance). HP ProLiant DL580 G7, (4 processors, 32 cores) 1,807,347 tpmC, .49/tpmC, available 10/15/10.

11 Best OLTP 2 CPU Performance34 Percent More Performance 32 Percent Less Cost Per Transaction Performance and Value Leadership Over IBM As of September 28, Source: Transaction Processing Performance Council (TPC), Oracle Database 11g Release 2 Standard Edition One and Oracle Linux on Cisco UCS c240 M3 High-Density Rack Server, 1,609, tpmC, $0.47/tpmC, available 9/27/12. IBM Power 780 Server Model 9179-MHB with IBM DB2 9.5, 1,200, tpmC, $0.69/tpmC, available 10/13/10.

12 Best Scalability and Performance Overall World Record SAP SD BenchmarkOracle’s Sun Fire X4800 M2 servers Highest Performance Ever! Near Linear Scaling SD Users As of September 28, Source: SAP AG, 2-Nodes: The SAP SD-Parallel standard application benchmark performed on September 26, 2001, by Oracle in Burlington, MA, USA, has been certified with the following data: 49,860 SAP SD-Parallel benchmark users seconds average dialog response time, 5,481,670 fully processed order line items per hour,  16,445,000 dialog steps per hour,  274,080 SAPS, seconds/0.030 seconds average dialog response time (dialog/update). CPU utilization of servers, 91% (node 1 active: 88%, node 2 active: 94%. Server configuration: 2 Sun Fire X4800 servers each with 8 processors / 80 cores / 160 threads, Intel Xeon Processor E7-8870, 2.40 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 30 MB L3 cache per processor, 512 GB main memory, Solaris 10 operating system, Oracle Database 11g Release 2 with Oracle Real Application Clusters, SAP enhancement package 4 for SAP ERP Certification #: 4-Nodes: The SAP SD-Parallel standard application benchmark performed on September 26, 2001, by Oracle in Burlington, MA, USA, has been certified with the following data: 94,736 SAP SD-Parallel benchmark users, 0.41 seconds average dialog response time, 10,921,000 fully processed order line items per hour.  32,763,000 dialog steps per hour.  546,050 SAPS, seconds/0.041 seconds average dialog response time (dialog/update). CPU utilization of servers, 93% (node 1 active: 89%, node 2 active: 94%. node 3 active: 94% node 4 active: 94%. Server configuration: 4 Sun Fire X4800 servers each with 8 processors / 80 cores / 160 threads, Intel Xeon Processor E7-8870, 2.40 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 30 MB L3 cache per processor, 512 GB main memory, Solaris 10 operating system, Oracle Database 11g Release 2 with Oracle Real Application Clusters, SAP enhancement package 4 for SAP ERP 6.0. Certification #: 6-Nodes: The SAP SD-Parallel standard application benchmark performed on September 20, 2011, by Oracle in Burlington, MA, has been certified with the following data: 137,904 SAP SD-Parallel benchmark users 0.81 seconds average dialog response time,  15,309,330 fully processed order line items per hour,  45,928,000 dialog steps per hour,  765,470 SAPS, seconds/0.045 seconds average dialog response time (dialog/update). CPU utilization of servers, 89% (node 1 active: 85%, node 2 active: 90%, node 3 active: 86%, node 4 active: 86%, node 5 active: 92%, node 6 active: 92%). Server configuration: 6 Sun Fire X4800 servers each with 8 processors / 80 cores / 160 threads, Intel Xeon Processor E7-8870, 2.40 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 30 MB L3 cache per processor, 512 GB main memory, Solaris 10 operating system, Oracle Database 11g Release 2 with Oracle Real Application Clusters, SAP enhancement package 4 for SAP ERP 6.0. Certification #: 8-Nodes: The SAP SD-Parallel standard application benchmark performed on September 5, 2011, by Oracle in Burlington, MA, has been certified with the following data: 180,000 SAP SD-Parallel benchmark users (Overall World Record SAP SD Benchmark result) seconds average dialog response time, 20,327,670 fully processed order line items per hour, 60,983,000 dialog steps per hour, 1,016,380 SAPS, seconds/0.055 seconds average dialog response time (dialog/update). CPU utilization of servers, 89% (node 1 active: 85%, node 2 active: 89%, node 3 active: 90%, node 4 active: 90%, node 5 active: 90%, node 6 active: 89%), node 8 active: 90%). Server configuration: 8 Sun Fire X4800 servers each with 8 processors / 80 cores / 160 threads, Intel Xeon Processor E7-8870, 2.40 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 30 MB L3 cache per processor, 512 GB main memory, Solaris 10 operating system, Oracle Database 11g Release 2 with Oracle Real Application Clusters, SAP enhancement package 4 for SAP ERP 6.0. Certification #: As of September 28, 2012 these results have been certified by SAP. Source: SAP AG: Please see notes page for SAP certification information for the above results.

13 Best Business Intelligence Performance SAP BI-Data Mart BenchmarkNear Linear Scaling 1,165,742 4-Node RAC Fujitsu RX300 3-Node RAC 2-Node RAC Single Node SMP As of September 28, 2012: Source: SAP AG, IBM DB2 (IBM DB2 Best SAP BI-D result) The SAP BI-D Standard Application Benchmark performed on October 17, 2008 by IBM in Rochester, MN, USA was certified on October 31, 2008 with the following data: Throughput/hour: 182,112 query navigation steps.CPU utilization of central system: 94% Operating system, Central server: i RDBMS: DB2 for i 6.1 Platform release: SAP NetWeaver 7.0 (2004s). Configuration: Central server: IBM Power System 570, 4 processors / 8 cores / 16 threads, POWER6, 5 GHz, 128 KB L1 cache and 4 MB L2 cache per core, 32 MB L3 cache per processor, 128 GB main memory. Certification number: Oracle SAP BI-D Results Single Node The SAP BI-D Standard Application Benchmark performed on August 8, 2009 by Fujitsu in Walldorf, Germany has been certified (certification number ) with the following data: Throughput/hour (query navigation steps): 320,363. Operating system, central server: SuSe Linux Enterprise Server 10. RDBMS: Oracle 10g Real Application Clusters (RAC). Technology platform release: SAP NetWeaver 7.0 (non-Unicode). Configuration: Central server: Fujitsu PRIMERGY RX300-S5, 2 processors / 8 cores / 16 threads, Intel Xeon Processor X5570, 2.93 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 8 MB L3 cache per processor, 96 GB main memory. Certification #: 2-Node The SAP BI-D Standard Application Benchmark performed on August 11, 2009 by Fujitsu in Walldorf, Germany has been certified (certification number ) with the following data: Throughput/hour (query navigation steps): 609,349. Operating system all servers: SuSE Linux Enterprise Server 10. RDBMS: Oracle 10g Real Application Clusters (RAC). Technology platform release: SAP NetWeaver 7.0 (non-Unicode). Configuration: 2 servers (2 active nodes): Fujitsu PRIMERGY RX300-S5, 2 processors / 8 cores / 16 threads, Intel Xeon Processor X5570, 2.93 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 8 MB L3 cache per processor, 96 GB main memory. Certification #: 3-Node The SAP BI-D Standard Application Benchmark performed on October 14, 2009 by Fujitsu in Walldorf, Germany has been certified (certification number ) with the following data: Throughput/hour (query navigation steps): 900,309. Operating system, all servers: SuSE Linux Enterprise Server 10. RDBMS: Oracle 10g Real Application Clusters (RAC). Technology platform release: SAP NetWeaver 7.0 (non-Unicode). Configuration: 3 servers (3 active nodes): Fujitsu PRIMERGY RX300-S5, 2 processors / 8 cores / 16 threads, Intel Xeon Processor X5570, 2.93 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 8 MB L3 cache per processor, 96 GB main memory. Certification #: 4-Node (World record SAP BI-D): The SAP BI-D Standard Application Benchmark performed on October by Fujitsu in Walldorf, Germany has been certified (certification number ) with the following data: Throughput/hour (query navigation steps): 1,165,742. Operating system all servers: SuSE Linux Enterprise Server 10. RDBMS: Oracle 10g Real Application Clusters (RAC). Technology platform release: SAP NetWeaver® 7.0 (non-Unicode). Configuration: 4 servers (4 active nodes): Fujitsu PRIMERGY RX300-S5, 2 processors / 8 cores / 16 threads, Intel Xeon Processor X5570, 2.93 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 8 MB L3 cache per processor, 96 GB main memory. Certification #: DB2 Oracle These results as of September 28, 2012: have been certified by SAP AG, Please see notes page for SAP benchmark certification details for the above results.

14 Best Business Intelligence Performance World Record SAP BI-Data Mart Benchmark2-Node RAC Fujitsu RX600 S5 As of September 28, 2012: Source: SAP AG, Best IBM DB2 SAP BI-Data Mart Result The SAP BI-D standard application benchmark performed on October 17, 2008 by IBM in Rochester, MN, USA was certified on October 31, 2008 with the following data: Throughput/hour: 182,112 query navigation steps, CPU utilization of central system: 94% . Operating system, Central server: i 6.1. RDBMS: DB2 for i 6.1. Technology platform release: SAP NetWeaver 7.0 (2004s)Configuration: Central server: IBM Power System 570, 4 processors / 8 cores / 16 threads, POWER6, 5 GHz, 128 KB L1 cache and 4 MB L2 cache per core, 32 MB L3 cache per processor, 128 GB main memory. Certification number: Single Node Oracle BI-Data Mart Result The SAP BI-D standard application benchmark performed on March 7, 2010 by Fujitsu in Walldorf, Germany, has been certified with the following data: Throughput/hour (query navigation steps): 854,649. CPU utilization of central server: 96% (1 node active: 96%). Operating System, central server: SuSE Linux Enterprise Server 10. RDBMS: Oracle 11g. Technology platform release: SAP NetWeaver® 7.0 technology platform (non-Unicode). Configuration: 1 server (1 active node): Fujitsu PRIMERGY RX600 S5, 4 processors / 32 cores / 64 threads, Intel Xeon Processor X7560, 2.26 GHz, 6.40GT/s Intel® QPI, 24 MB L3 cache per processor, 128 GB main memory. Certification number: 2-Node Oracle BI-Data Mart World Record Result The SAP Business Intelligence-Data Mart (BI-D) standard application benchmark performed on May, 6, 2010 by Fujitsu in Walldorf, Germany has been certified with the following data: Throughput/hour (query navigation steps): 1,624,629. CPU utilization of central server: 94% (2 nodes active: 93% / 94%). Operating system, central server: SuSE Linux Enterprise Server 10. RDBMS: Oracle Database 11g . Technology platform release: SAP NetWeaver® 7.0 technology platform (non-Unicode). Configuration: 2 server (2 active nodes), each: Fujitsu PRIMERGY RX600 S5, 4 processors / 32 cores / 64 threads, Intel Xeon Processor X7560, 2.26 GHz, 6.40GT/s Intel® QPI , 24 MB L3 cache per processor, 128 GB main memory. Certification number: These results as of September 28, 2012 have been certified by SAP AG, Please see notes page for SAP benchmark certification details for the above results.

15 High Performance TodayDatabase Scaling Storage

16 Oracle Beehive OLTP on ExadataAustin Data Center Utah Data Center Data Guard Primary Standby Runs Oracle , Calendar, Contacts, Chat, Documents, Web Conferencing 9 X2-2 DB Machine 2376 cores, 2PB storage, 7TB DRAM, 48 TB Flash Complete Oracle Software Stack RAC, Streams, Active Data Guard, Secure Backup, RMAN, Flashback Database, ASM, Partitioning 2X space saved with compressed SecureFiles Dev / Test

17 Turkcell: DW and DB ConsolidationBenefits Reduced Admin 20% Storage Savings 900 TB 1,000 TB to 100 TB Faster Reports 10X 27 min to 3 min (avg for 50k rpts) “In a word, Oracle Exadata is fantastic. Almost no report takes more than 10 minutes to run, versus hours before. It sounds unreal, but it’s real.” - Power User, Finance Department, Turkcell 80% Less Power 30 m2 Less Space Data Center Cost Savings Objectives Speed up BI Lean, green data center Prepare for big data growth Solution 2010: Replace 11 racks with 1 full-rack Exadata V2 for DW 2011: Add 2 full-rack Exadata X2-2s for DB consolidation Pre-Exadata Data Warehouse Exadata V2 Data Warehouse 2 Exadata X2-2 Prod Original V2 Prod/DR/Dev Hitachi USP-V 5 Racks EMC DMX-4 5 Racks 2010 2011 Backup Restore Turkcell is the largest mobile phone operator in Turkey, with twice as many subscribers as the nearest competitor. In 2010, Turkcell’s business intelligence agility was throttled by the performance of its data warehouse platform. Using a large Unix SMP system and 10 racks of high-end storage, the average report took almost 30 minutes to complete, and 50,000 reports were being run monthly. That’s when Turkcell invested in a full rack Exadata Database Machine and, applying Exadata’s hybrid columnar compression, converted a 250 TB data warehouse into a 25 TB compressed data warehouse on Exadata, enjoying a 10x speedup in the average report, which completed in less than 3 minutes. In 2011, Turkcell decided to consolidate more databases onto Exadata, and invested in two full rack Exadata X2-2 systems, holding 600 TB of raw data compressed to 60 TB. The original Exadata system now holds 400 TB of raw data compressed to 40 TB, for a total of 1 Petabyte of raw data on Exadata. In addition to the 10x performance improvement in their reporting, Turkcell freed up 900 TB of tier-1 storage for other uses, reduced overall IT admin costs by 20%, and shrunk the footprint of their data center, including power consumption, by 80%. 2011 250 TB Raw Data 25 TB Compressed 4 Prod, 2 Test Databases 400 TB Raw/ 40 TB Compressed 2 RAC Clusters 2 Prod Databases 600 TB Raw / 60 TB Compressed 16-node RAC Cluster

18 In-Memory & Parallelismfor High Performance

19 Oracle TimesTen In-Memory Database CacheTelco Services Financial Services In-Memory Database Cache Application Real-Time Analytics – Dashboard, Scorecard Data Mart eCommerce, Personalization Deploys in the application tier Caches subset of Oracle Database Full featured in-memory RDBMS Standard SQL and PL/SQL Accelerates applications with micro-second response time Built-in high availability Read/write caching Automatic synchronization with Oracle Database

20 Lightning Fast Response TimeOracle TimesTen In-Memory Database Intel Xeon Ghz, 2 CPUs, 6 cores/CPU - Oracle Linux 5.6 7.0 1.78

21 TimesTen Read ThroughputOver 4.14M Reads per Sec Oracle TimesTen In-Memory Database Intel Xeon Ghz, 2 CPUs, 6 cores/CPU - Oracle Linux 5.6

22 TimesTen Update Throughput647K Updates per Sec Oracle TimesTen In-Memory Database Intel Xeon Ghz, 2 CPUs, 6 cores/CPU - Oracle Linux 5.6

23 12.1 In-Memory Parallel ExecutionHigh performance parallel query processing on memory cached data Queries run from tables in database buffer cache Harnesses memory capacity of entire database cluster for queries An affinity algorithm places fragments of a object (partitions) in memory on different RAC nodes Data is kept compressed (& columnar) in memory Memory has 100x the bandwidth of disk

24 12.1 Adaptive Execution Plans Good SQL execution without interventionTable scan Order _items NESTED LOOPS Index Scan Prod_info_ind HASH JOIN Prod_info Statistics Collector Threshold exceeded subplan switches Plan decision deferred until runtime Final decision is based on statistics collected during execution If statistics prove to be out of range, sub-plans can be swapped Eliminates suboptimal plans caused by data skew Statistics Collector Plan decision deferred until runtime Final decision is based on statistics collected during execution Alternate sub-plans are pre-computed, and stored in the cursor Statistic collectors are inserted at key points in the plan Each sub-plan has a valid range for stats collected If stats prove to be out of range sub-plans can be swapped Requires buffering near the swap point to avoid returning rows to user Only join methods and distribution method can change

25 12.1 Adaptive Execution Plans Good SQL execution without interventionHJ Plan decision deferred until runtime Final decision is based on statistics collected during execution If statistics prove to be out of range, sub-plans can be swapped Bad effects of skew eliminated Table scan T1 HJ T2 Table scan T1 NL Index Scan T2 Threshold exceeded, plan switches Table scan T2 Plan decision deferred until runtime Final decision is based on statistics collected during execution Alternate sub-plans are pre-computed, and stored in the cursor Statistic collectors are inserted at key points in the plan Each sub-plan has a valid range for stats collected If stats prove to be out of range sub-plans can be swapped Requires buffering near the swap point to avoid returning rows to user Only join methods and distribution method can change

26 12.1 PL/SQL “with” execution within SQL8x faster SQL using New runtime & “With functions” Trades Memory for Performance PL/SQL functions defined inside scope of a SQL Saving on the CPU costs of row by row transitions & data passing with  function is_number(s varchar2) return char is  begin     if to_number(s) is not null then        return 'YES'; …  end; select count(*) from user_requests where is_number(quantity) = 'NO'

27 12.1 Fast Safe Callouts Speeds up Oracle R Enterprise, now built into the Oracle DatabaseR is integrated into SQL R workspace console BI, Workflows, Web Services In-database statistical & predictive techniques Transparent access via function push-down Embedded R engine Trades off memory for speed, array processing Safe callouts up to 20x faster Enables better performance for several User Defined Function based R statistical functions select max(rqGamma(ARRDELAY)) A  from ONTIME_S group  by YEAR

28 12.1 In-Memory Global Temporary TablesOperations for temporary tables without redo or undo I/O Global Temporary Tables (GTT) are frequently used in OLTP and DW e.g. staging of intermediate results in reports Starting 12.1, GTTs run purely in-memory They do not generate I/O for redo, undo Transaction-duration GTT avoid generating data or index I/O as well if the transaction commits before DBWR writes the blocks out Enables use of Global Temporary Tables on Active Data Guard Reduced WAN bandwidth to remote Standby Faster Database Recovery Use Case Redo in 11g Redo in 12c OLTP on GTT 10,000 transactions 447MB ZERO Internal use of Temporary Tablespace for sorts etc already is no logging Global Temporary Tables are heavily used in OLTP and DW For staging intermediate results in Reports & Transactions; GTTs often joined back to other tables using SQL for additional processing All data in GTTs is private within a session and is deleted when the transaction or session ends New GTT implementation does not generate redo Improves performance of DMLs on GTT Enables Applications to make intensive use of GTT without worrying about redo No Logging GTTs Enables Reporting on Active Data Guard Currently Oracle Applications reports use GTT GTTs currently generate redo, creates a barrier to using ADG No Logging GTTs make ADG usable for Oracle Apps & other OLTP applications

29 12.1 In-Memory LOB Queries & UpdatesSpeeds up string ops on LOB & updates of Temporary LOBs In-Memory optimization to trade PGA memory for speed Uses in-memory working area for Temporary LOBs that are small Automatically and transparently spills temporary LOB to disk / Flash Cache as LOB grows beyond a threshold Speeds up all LOB string operations concatenate, append, substr, length, instr, compare, trim, like, replace, pad, nvl using SQL functions or DBMS_LOB package

30 12.1 XML faster, more parallelStructured XML All SQL/XML operators for XML generation are PQ enabled Semistructured XML XMLTable operator is orders of magnitude faster Non Schema based XML query and index optimizations TPoX is 2x faster than (XQuery, XQuery Update, non-schema based XML)

31 12.1 Parallel Load & Move LOBs5x to 17x speedup in loading & moving SecureFiles Enhanced Parallelism DML, CTAS & MOVE for SecureFiles Intra-partition parallelism Parallel move for Non-partitioned table Linear Scaling with degree of parallelism Helps exploit multi-core & I/O parallelism

32 12.1 Sharded AQ-JMS Queues Support for greater parallelismSingle Consumer, Single Instance Optimized for high concurrency and large number of subscribers in RAC Optimized for the JMS standard Backwards Compatible for Standard JMS based applications just recreate the AQ in the database Multconsumer, Two Node RAC

33 12.1 Oracle Spatial and Graph with CBOQuery execution time reduced by 10x to 100x Main Memory Improved Oracle Spatial and Graph Extensible Optimizer performs two important functions Statistics Collection Histogram construction using various spatial heuristics e.g. Equal number of data rectangles in each bucket or buckets with equal areas Sampling is ineffective for spatial data but we provide a fast main memory-based algorithm Selectivity and Cost Estimation Dependent on collected statistical information . . . A I L G H E B C D K J F

34 12.1 In-Memory Queue based Fast AuditingUnified Audit Trail and Unified Audit Configuration Use of in-memory queues to avoid overhead on user transaction enables auditing to be enabled on all apps Audit queues are persisted using SecureFile store Efficient management and Cleanup of audit trail Partitioned audit trail for faster cleanup Size based partitions to store audit trail data Conditional audit support for selective auditing

35 In-memory scaling with Pluggable Databases Highly Efficient: 6x Less H/W Resource, 5x more ScalableOLTP benchmark comparison Only 3GB of memory vs. 20GB memory used for 50 databases Pluggable databases scaled to over 100 while separate database instances maxed at 50 Memory Utilized # Databases

36 High-Performance with Large Data Volumes

37 Data Growth ChallengesIT must support exponentially growing amounts of data With improved performance With lower cost Powerful and efficient compression is key

38 12.1 Technology Advances for VLDBOracle Database 12c features for Compression & ILM Advanced Row Compression Optimizations Faster queries on compressed data Heat Map Insight into Data usage to help optimize for price/performance Automatic Data Optimization Automatic compression & storage tiering RMAN read-only optimization for Backup & Recovery In-DB Archiving

39 12.1 Advanced Row CompressionQueries on Compressed Data In-memory Scans Data is never expanded into uncompressed form in memory In 12.1, up to 3x faster In-memory Scans for Complex predicates on Columns with small number of distinct values select … from Customers where name like ‘%C%Bank’

40 Real World Compression Results - ERP Database 10 Largest TablesStorage Utilization MB 3x Smaller Table Scan Performance (seconds) Time 2.5x Faster DML Performance Less than 3% Overhead

41 12.1 Heat Map – visualize use of dataInsight helps identify opportunity & drive automation Table Level Heat Map Partition Level Heat Map Row Level Heat Map

42 12.1 Automatic Data OptimizationAs data ages: Activity declines Volume grows Older data primarily for reporting Compliance & Reporting Reporting OLTP 10x compressed 15x compressed This Quarter This Year Prior Years alter table … add policy … compress for query after 3 months of no modification … compress for archive after 1 year … Row Store for fast OLTP Compressed Column Store for fast analytics Archive Compressed Column Store for max compression As data cools down, Advanced Data Optimization automatically converts data to columnar compressed Online

43 12.1 Automatic Data OptimizationEnables Columnar & In-Memory in OLTP Automatic Compression & Storage Tiering Users can customize policies & scheduling 100% Online, Background operation Optimized use of Row Store & Column Store within a single table Enables use of Column Store in OLTP environments ADO & 12.1 HCC Row Level Locking Enables greater In-Memory Processing 3x Row Compression, 10x Query Compression & 15x or more Archive Compression help more of the data fit in DRAM & Flash Memory Row Store gives Fast Loads & Updates Column Store enables fast Analytics & Reporting

44 12.1 In-Database Archive 5x speedup for upgrades and reportsApplications typically work with recent data But often need to retain data for 5 to 10 years In-DB Archiving feature provides the ability to archive infrequently used data within the database Archived Data is invisible by default Speedup up upgrades & Reports by up to 5x Works with Partition Pruning, and Exadata Storage Index to eliminate I/O for archived data Archived data remains online for SQL Query & DMLs & is upgraded with the App Easily enabled for a table: alter table … row archival Application can marks rows as archived: update SALES_ORDERS … set ORA_ARCHIVE_STATE = 1 Sessions can set default visibility to see all data or active data only (default) alter session set row archival visibility = [all| active]

45 12.1 ADO for VLDB Backup & RecoveryAs data ages: It becomes read- mostly need not be backed up repeatedly Compliance & Reporting Reporting OLTP 10x compressed 15x compressed alter table add policy … move to tbs_archive readonly Read / Write Tablespace Read-mostly data is moved to a READONLY Tablespace READONLY TBS As data cools down, Advanced Data Optimization automatically moves it to a READONLY TBS, its backed up only once after that

46 Advanced Compression – Smaller & Faster100% Application Transparent End-to-end Cost/Performance Benefits across CPU, DRAM, Flash, Disk & Network Runs Faster: OLTP Apps (both Transactional & Analytics) & DW Greater speedup from In-memory (3-10x more fits in Buffer Cache & Flash Cache) Faster Loads (Load into uncompressed, then ADO in background) Faster Queries (Query-on-compresed, Columnar in OLTP) Faster Backup & Restore Speeds Reduces Footprint CapEx & OpEx savings – current & ongoing CapEx: Lowers Server & Storage Cost for Primary, Standby, Backup, Test & Dev Databases OpEx: Lowers Heating, Cooling, Floor space Costs, Admin (declarative) Up to 10x greater Consolidation on the same hardware Increases Cloud ROI through Database Footprint reduction in DRAM Memory

47 End-to-end Performance Architecture

48 Exadata X3 Database In-Memory MachineExtreme Performance 100 GB/s Scan speed over Uncompressed Row Store 27 TB/hr Backup speed to disk 16 TB/hr Load speed 1.5 Million Read IOPs and 1 Million Write IOPs Column Store with average 10x Query Compression and 15x Archive Compression; Customers report up to 50x

49 Summary SMP Scale-Up and RAC Scale-OutWorld Record holder in database performance Real World High Performance Database environments In-Memory and Flash for High Performance High Performance with Large Data Volume End-to-end Performance Architecture

50

51