Oracle Database Performance Tuning: The Not SQL Option
1 like2,241 views
This document appears to be a slide presentation by Daniel A. Morgan on non-SQL performance tuning. It provides biographical details on Morgan, including that he is an Oracle ACE Director, author, and consultant. It also lists locations where Morgan has presented, including various universities and OTN events in Asia Pacific, Europe, Middle East, Africa, and Latin America. The presentation focuses on performance tuning methodology and addressing the root causes of issues rather than just symptoms.
![Detect NUMA Use
[oracle@alpha1 ~]$ numactl --hardware
available: 1 nodes (0)
node 0 size: 4095 MB
node 9 free: 648MB
node distances:
node 0
0: 10
[oracle@alpha1 ~]$ numactl --show
policy: default
preferred node: current
physcpubind: 0
nodebind: 0
membind: 0
[oracle@alpha1 ~]$ numactl --hardware
available: 1 nodes (0)
node 0 size: 4095 MB
node 9 free: 648MB
node distances:
node 0
0: 10
[oracle@alpha1 ~]$ numactl --show
policy: default
preferred node: current
physcpubind: 0
nodebind: 0
membind: 0
[dmorgan@lxorap1n5 ~]$ numactl --hardware
available: 2 nodes (0-1)
NUMA Not Configured
NUMA Configured
Daniel A. Morgan | damorgan12c@gmail.com | www.morganslibrary.org
Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015
23
available: 2 nodes (0-1)
node 0 size: 48457 MB
node 0 free: 269 MB
node 1 size: 48480 MB
node 1 free: 47 MB
node distances:
node 0 1
0: 10 20
1: 20 10
[dmorgan@lxorap1n5 ~]$ numactl --show
policy: default
preferred node: current
physcpubind: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
cpubind: 0 1
nodebind: 0 1
membind: 0 1](https://image.slidesharecdn.com/guatemala15notsql-150819031415-lva1-app6891/85/Oracle-Database-Performance-Tuning-The-Not-SQL-Option-23-320.jpg)
Oracle Database Performance Tuning: The Not SQL Option
- 1. Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 1 Oracle Database Performance Tuning: The Not SQL Option Daniel A. Morgan 17 August, 2015
- 2. Daniel A. Morgan Oracle ACE Director Educator Author and Primary Instructor: University of Washington Oracle program Oracle Consultant: Harvard University University Guest Lecturers APAC: University of Canterbury (NZ) EMEA: University of Oslo (Norway) Latin America: Universidad Latina de Panama and Technologico de Costa Rica Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 2 Latin America: Universidad Latina de Panama and Technologico de Costa Rica The Morgan behind Morgan's Library on the web www.morganslibrary.org 10g, 11g, and 12c Beta tester Co-Founder International GoldenGate Oracle Users Group Principal Advisor: Forsythe Meta7
- 3. The Morgan's Library Web Site Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 3
- 4. Travel Log: OTN Latin America Tour 2009 Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 4
- 5. Travel Log: OTN Latin America Tour 2010 Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 5
- 6. Travel Log: OTN Latin America Tour 2014 Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 6
- 7. Travel Log: OTN Latin America Tour 2014 Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 7
- 8. $ cd $MORGAN_HOME Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 8 The most important thing I discovered during this trip is the how disruptive our presence was to the locals.
- 9. Content Density Warning Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 9 The most important thing I discovered during this trip is the how disruptive our presence was to the locals. Take Notes ... Ask Questions
- 10. The #1 Database Performance Tuning Mythology Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 10 This was from the worst performing Oracle Database I've seen since 2006
- 11. The #1 Database Performance Tuning Methodology Guessing Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 11
- 12. One Solution Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 12
- 13. BAAG Comrades Christian Antognini Karl Arao Mark Bobak Ronald Bradford Wolfgang Breitling Doug Burns Andrew Clarke Randolf Geist Alex Gorbachev Marco Gralike BAAG Membership Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 13 Marco Gralike Frits Hoogland John Hurley Anjo Kolk David Kurtz Jonathan Lewis Robyn Sands Jared Still Jeremiah Wilton
- 14. Clearly No surprise ... I do not endorse guessing Possibly a big surprise ... I do not like normal AWR reports I think the solutions to every performance issue is an Exadata And I am not attracted to all the tools with pretty GUIs So let us take a deep dive into performance tuning and address the root cause of the majority of issues I see in my work We will focus on what fixes all issues ... not just one issue Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 14 We will focus on what fixes all issues ... not just one issue We all know that 30+ years of doing it the way we have been DBMS_SUPPORT introduced with version 7.2 DBMS_TRACE introduced with version 8.1.5 DBMS_MONITOR introduced with version 10.1 has not eliminated tuning problems
- 15. What Affects Performance Hardware Servers Resources CPU Memory Bus Bandwidth and Latency Storage Subsystems Networks Software Let's focus on what is importantLet's focus on what is important Application Web Servers Application Servers Middleware Caching Application Code Quality Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 15 Operating System Configuration Virtual Machines Drivers Database Memory Allocation Optimizer Configuration SQL Quality Let's focus on what is important but not on the radar Let's focus on what is important but not on the radar
- 16. Hardware Servers and Operating Systems Blade Servers I/O Cards NUMA Architecture HugePages Swapiness Virtual Machines Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 16 Virtual Machines Storage Controllers Read-Write Caches LUN Size and Layout Networks TCP/IP UDP
- 17. Blade Servers (1:2) Suitable for web and application servers Possibly usable for small databases A meltdown waiting to happen with RAC RAC: Frequently Asked Questions (Doc ID 220970.1) Cluster interconnect network separation can be satisfied either by using standalone, dedicated switches, which provide the highest degree of network isolation, or Virtual Local Area Networks defined on the Ethernet switch, which provide broadcast domain isolation between IP networks. VLANs are fully RAC: Frequently Asked Questions (Doc ID 220970.1) Cluster interconnect network separation can be satisfied either by using standalone, dedicated switches, which provide the highest degree of network isolation, or Virtual Local Area Networks defined on the Ethernet switch, which provide broadcast domain isolation between IP networks. VLANs are fully Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 17 isolation, or Virtual Local Area Networks defined on the Ethernet switch, which provide broadcast domain isolation between IP networks. VLANs are fully supported for Oracle Clusterware interconnect deployments. Partitioning the Ethernet switch with VLANs allows for: - Sharing the same switch for private and public communication. - Sharing the same switch for the private communication of more than one cluster. - Sharing the same switch for private communication and shared storage access. The following best practices should be followed: The Cluster Interconnect VLAN must be on a non-routed IP subnet. All Cluster Interconnect networks must be configured with non-routed IPs. The server-server communication should be single hop through the switch via the interconnect VLAN. There is no VLAN-VLAN communication. Oracle recommends maintaining a 1:1 mapping of subnet to VLAN. The most common VLAN deployments maintain a 1:1 mapping of subnet to VLAN. It is strongly recommended to avoid multi-subnet mapping to a single VLAN. Best practice recommends a single access VLAN port configured on the switch for the cluster interconnect VLAN. The server side network interface should have access to a single VLAN. isolation, or Virtual Local Area Networks defined on the Ethernet switch, which provide broadcast domain isolation between IP networks. VLANs are fully supported for Oracle Clusterware interconnect deployments. Partitioning the Ethernet switch with VLANs allows for: - Sharing the same switch for private and public communication. - Sharing the same switch for the private communication of more than one cluster. - Sharing the same switch for private communication and shared storage access. The following best practices should be followed: The Cluster Interconnect VLAN must be on a non-routed IP subnet. All Cluster Interconnect networks must be configured with non-routed IPs. The server-server communication should be single hop through the switch via the interconnect VLAN. There is no VLAN-VLAN communication. Oracle recommends maintaining a 1:1 mapping of subnet to VLAN. The most common VLAN deployments maintain a 1:1 mapping of subnet to VLAN. It is strongly recommended to avoid multi-subnet mapping to a single VLAN. Best practice recommends a single access VLAN port configured on the switch for the cluster interconnect VLAN. The server side network interface should have access to a single VLAN.
- 18. Blade Servers (2:2) Blade servers, of which Cisco UCS is but one example, do not have sufficient independent network cards to avoid the networking becoming a single point of failure It is good when the public interface has a "keep alive" enabled but this is a fatal flaw for the cluster interconnect as fail-over will be delayed When different types of packets, public, storage, and interconnect are mixed low-level diagnostics are difficult ... if not impossible When different types of packets, public, storage, and interconnect are Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 18 When different types of packets, public, storage, and interconnect are mixed the latency of one is the latency of all Traffic from any one blade can impact all blades Troubleshooting gc block lost and Poor Network Performance in a RAC Environment (Doc ID 563566.1) 6. Interconnect LAN non-dedicated Description: Shared public IP traffic and/or shared NAS IP traffic, configured on the interconnect LAN will result in degraded application performance, network congestion and, in extreme cases, global cache block loss. Action: The interconnect/clusterware traffic should be on a dedicated LAN defined by a non-routed subnet. Interconnect traffic should be isolated to the adjacent switch(es), e.g. interconnect traffic should not extend beyond the access layer switch(es) to which the links are attached. The interconnect traffic should not be shared with public or NAS traffic. If Virtual LANs (VLANS) are used, the interconnect should be on a single, dedicated VLAN mapped to a dedicated, non-routed subnet, which is isolated from public or NAS traffic. Troubleshooting gc block lost and Poor Network Performance in a RAC Environment (Doc ID 563566.1) 6. Interconnect LAN non-dedicated Description: Shared public IP traffic and/or shared NAS IP traffic, configured on the interconnect LAN will result in degraded application performance, network congestion and, in extreme cases, global cache block loss. Action: The interconnect/clusterware traffic should be on a dedicated LAN defined by a non-routed subnet. Interconnect traffic should be isolated to the adjacent switch(es), e.g. interconnect traffic should not extend beyond the access layer switch(es) to which the links are attached. The interconnect traffic should not be shared with public or NAS traffic. If Virtual LANs (VLANS) are used, the interconnect should be on a single, dedicated VLAN mapped to a dedicated, non-routed subnet, which is isolated from public or NAS traffic.
- 19. I/O (1:3) Not all HBA cards are the same NIC cards vary widely in capabilities and performance TCP/IP Off-loading Kernel Optimization of the TCP/IP stack --enable TCP kernel auto-tuning /proc/sys/net/ipv4/tcp_moderate_rcvbuf (1=on) -- tune TCP max memory: tune to 2xBDP (Bandwidth x Delay Product) -- For example, with 40 Mbits/sec bandwidth, 25 msec delay, --enable TCP kernel auto-tuning /proc/sys/net/ipv4/tcp_moderate_rcvbuf (1=on) -- tune TCP max memory: tune to 2xBDP (Bandwidth x Delay Product) -- For example, with 40 Mbits/sec bandwidth, 25 msec delay, Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 19 -- For example, with 40 Mbits/sec bandwidth, 25 msec delay, -- BDP = (40 x 1000 / 8 Kbytes/sec) x (0.025 sec) ~ 128 Kbytes /proc/sys/net/ipv4/tcp_rmem and tcp_wmem 4096 87380 174760 -- tune the socket buffer sizes by setting to 2xBDP /proc/sys/net/core/rmem_max and wmem_max -- ensure that TCP Performance features are enabled /proc/sys/net/ipv4/tcp_sack /proc/sys/net/ipv4/tcp_window_scaling /proc/sys/net/ipv4/tcp_timestamps -- additionally be sure NIC cards have TCP off-loading capability -- For example, with 40 Mbits/sec bandwidth, 25 msec delay, -- BDP = (40 x 1000 / 8 Kbytes/sec) x (0.025 sec) ~ 128 Kbytes /proc/sys/net/ipv4/tcp_rmem and tcp_wmem 4096 87380 174760 -- tune the socket buffer sizes by setting to 2xBDP /proc/sys/net/core/rmem_max and wmem_max -- ensure that TCP Performance features are enabled /proc/sys/net/ipv4/tcp_sack /proc/sys/net/ipv4/tcp_window_scaling /proc/sys/net/ipv4/tcp_timestamps -- additionally be sure NIC cards have TCP off-loading capability
- 20. I/O (2:3) Optimize SQL*Net --sqlnet.ora NAMES.DIRECTORY_PATH= (TNSNAMES, EZCONNECT) DEFAULT_SDU_SIZE=32767 -- listener.ora DGLOGSHIPB = (DESCRIPTION = (SDU = 32767) (SEND_BUF_SIZE=9375000) (RECV_BUF_SIZE=9375000) (ADDRESS_LIST = (ADDRESS = (PROTOCOL = TCP)(HOST = 10.0.7.2)(PORT = 1526)) ) --sqlnet.ora NAMES.DIRECTORY_PATH= (TNSNAMES, EZCONNECT) DEFAULT_SDU_SIZE=32767 -- listener.ora DGLOGSHIPB = (DESCRIPTION = (SDU = 32767) (SEND_BUF_SIZE=9375000) (RECV_BUF_SIZE=9375000) (ADDRESS_LIST = (ADDRESS = (PROTOCOL = TCP)(HOST = 10.0.7.2)(PORT = 1526)) ) Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 20 ) (CONNECT_DATA = (SERVICE_NAME = prodb) ) ) ) (CONNECT_DATA = (SERVICE_NAME = prodb) ) )
- 21. I/O (3:3) Optimize for RAC Read the Oracle installation documents with very careful attention to the advice given for kernel parameters If on Linux and you don't know what rmem and wmem are ... read the docs If on Solaris and you don't know what rsize and wsize are ... read the docs Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 21
- 22. NUMA Memory Allocation Non-Uniform Memory Access A memory design used in multiprocessing, where the memory access time depends on the memory location relative to the processor A processor can access its own local memory faster than non-local memory The benefits of NUMA are limited to particular workloads, notably on servers where the data are often associated strongly with certain tasks or users Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 22 tasks or users Diagram Source: Wikipedia
- 23. Detect NUMA Use [oracle@alpha1 ~]$ numactl --hardware available: 1 nodes (0) node 0 size: 4095 MB node 9 free: 648MB node distances: node 0 0: 10 [oracle@alpha1 ~]$ numactl --show policy: default preferred node: current physcpubind: 0 nodebind: 0 membind: 0 [oracle@alpha1 ~]$ numactl --hardware available: 1 nodes (0) node 0 size: 4095 MB node 9 free: 648MB node distances: node 0 0: 10 [oracle@alpha1 ~]$ numactl --show policy: default preferred node: current physcpubind: 0 nodebind: 0 membind: 0 [dmorgan@lxorap1n5 ~]$ numactl --hardware available: 2 nodes (0-1) NUMA Not Configured NUMA Configured Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 23 available: 2 nodes (0-1) node 0 size: 48457 MB node 0 free: 269 MB node 1 size: 48480 MB node 1 free: 47 MB node distances: node 0 1 0: 10 20 1: 20 10 [dmorgan@lxorap1n5 ~]$ numactl --show policy: default preferred node: current physcpubind: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 cpubind: 0 1 nodebind: 0 1 membind: 0 1
- 24. Is Your Database NUMA Aware? SQL> SELECT a.ksppinm PNAME, c.ksppstvl PVAL, a.ksppdesc PDESC 2 FROM x$ksppi a, x$ksppcv b, x$ksppsv c 3 WHERE a.indx = b.indx 4 AND a.indx = c.indx 5 AND LOWER(a.ksppinm) LIKE '%numa%' 6* ORDER BY 1; PNAME PVAL PDESC -------------------------- -------------- --------------------------------------------------------------- _NUMA_instance_mapping Not specified Set of nodes that this instance should run on _NUMA_pool_size Not specified aggregate size in bytes of NUMA pool _db_block_numa 1 Number of NUMA nodes _enable_NUMA_interleave TRUE Enable NUMA interleave mode _enable_NUMA_optimization FALSE Enable NUMA specific optimizations _enable_NUMA_support FALSE Enable NUMA support and optimizations SQL> SELECT a.ksppinm PNAME, c.ksppstvl PVAL, a.ksppdesc PDESC 2 FROM x$ksppi a, x$ksppcv b, x$ksppsv c 3 WHERE a.indx = b.indx 4 AND a.indx = c.indx 5 AND LOWER(a.ksppinm) LIKE '%numa%' 6* ORDER BY 1; PNAME PVAL PDESC -------------------------- -------------- --------------------------------------------------------------- _NUMA_instance_mapping Not specified Set of nodes that this instance should run on _NUMA_pool_size Not specified aggregate size in bytes of NUMA pool _db_block_numa 1 Number of NUMA nodes _enable_NUMA_interleave TRUE Enable NUMA interleave mode _enable_NUMA_optimization FALSE Enable NUMA specific optimizations _enable_NUMA_support FALSE Enable NUMA support and optimizations Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 24 _enable_NUMA_support FALSE Enable NUMA support and optimizations _numa_buffer_cache_stats 0 Configure NUMA buffer cache stats _numa_shift_enabled TRUE Enable NUMA shift _numa_shift_value 0 user defined value for numa nodes shift _numa_trace_level 0 numa trace event _px_numa_stealing_enabled TRUE enable/disable PQ granule stealing across NUMA nodes _px_numa_support_enabled FALSE enable/disable PQ NUMA support _rm_numa_sched_enable FALSE Is Resource Manager (RM) related NUMA scheduled policy enabled _rm_numa_simulation_cpus 0 number of cpus for each pg for numa simulation in resource mgr _rm_numa_simulation_pgs 0 number of PGs for numa simulation in resource manager _enable_NUMA_support FALSE Enable NUMA support and optimizations _numa_buffer_cache_stats 0 Configure NUMA buffer cache stats _numa_shift_enabled TRUE Enable NUMA shift _numa_shift_value 0 user defined value for numa nodes shift _numa_trace_level 0 numa trace event _px_numa_stealing_enabled TRUE enable/disable PQ granule stealing across NUMA nodes _px_numa_support_enabled FALSE enable/disable PQ NUMA support _rm_numa_sched_enable FALSE Is Resource Manager (RM) related NUMA scheduled policy enabled _rm_numa_simulation_cpus 0 number of cpus for each pg for numa simulation in resource mgr _rm_numa_simulation_pgs 0 number of PGs for numa simulation in resource manager
- 25. Enable Database NUMA Support conn / as sysdba ALTER SYSTEM SET "_enable_NUMA_support" = TRUE COMMENT= 'NUMA Support Enabled 15-Mar-2015' CONTAINER=ALL SCOPE=SPFILE SID='*'; conn / as sysdba ALTER SYSTEM SET "_enable_NUMA_support" = TRUE COMMENT= 'NUMA Support Enabled 15-Mar-2015' CONTAINER=ALL SCOPE=SPFILE SID='*'; Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 25
- 26. Also known as "Large Memory Pages" or just "Large Pages" Each page table entry represents a “virtual to physical” translation of a process’s memory Can be as large as 64 KB in size per entry Can be huge for large memory systems See PageTables in /proc/meminfo As large a 1.5 GB HugePages Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 26 As large a 1.5 GB The entire SGA must fit inside the HugePages. If it does not fit none of it will use the HugePage memory and you will essentially have walled your database off from using a large portion of the server's memory
- 27. Swapping (aka Paging) In a sense the operating system's version of the Oracle Temp tablespace Swapiness Value Strategy vm.swapiness=0 The kernel will swap only to avoid an out of memory condition vm.swapiness=60 The default value vm.swapiness=100 The kernel will swap aggressively Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 27
- 28. Leave sufficient cpu resources for the bare-metal operating system to perform I/O and manage network traffic Disable interrupt coalescing Disable chipset power management Read the docs http://www.vmware.com/pdf/Perf_Best_Practices_vSphere5.0.pdf VMs on NUMA machines should be configured to enhance Virtual Machines Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 28 VMs on NUMA machines should be configured to enhance memory allocation This example is from vSphere where 0 and 1 are the processor sockets numa.nodeAffinity=0,1numa.nodeAffinity=0,1
- 29. Storage Hardware Storage Spinning Disk Solid State Devices Controllers SAN Switches Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 29
- 30. Storage Heat Map Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 30
- 31. Storage Layout LUN 204 LUN 208 LUN 74 LUN 101 LUN 57 LUN 123 LUN 92 LUN 62 LUN 106 LUN 107 LUN 51 LUN 102 LUN 105 LUN 58 LUN 47 LUN 126 LUN 103 LUN 85 LUN 206 LUN 44 LUN 56 LUN 45 LUN 67 LUN 14 LUN 17 LUN 26 LUN 8 LUN 2 LUN 4 LUN c0vmware01p, c0vmware...c0vmware01p, c0vmware... c0vmware02p, c0vmware...c0vmware02p, c0vmware... c0ora01pc0ora01p c0odsrac01p, c0odsrac02pc0odsrac01p, c0odsrac02p c0ods04pc0ods04p c0orarac01p, c0orarac...c0orarac01p, c0orarac... c0visdb03p, c0visdb04pc0visdb03p, c0visdb04p c0orademand01pc0orademand01p LUN 204 LUN 208 LUN 74 LUN 101 LUN 57 LUN 123 LUN 92 LUN 62 LUN 106 LUN 107 LUN 51 LUN 102 LUN 105 LUN 58 LUN 47 LUN 126 LUN 103 LUN 85 LUN 206 LUN 44 LUN 56 LUN 45 LUN 67 LUN 14 LUN 17 LUN 26 LUN 8 LUN 2 LUN 4 LUN c0vmware01p, c0vmware... c0vmware02p, c0vmware... c0ora01p c0odsrac01p, c0odsrac02p c0ods04p c0orarac01p, c0orarac... c0visdb03p, c0visdb04p c0orademand01p Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 31 LUN 200 LUN 205 LUN 94 LUN 81 LUN 78 LUN 82 LUN 77 LUN 35 LUN 49 LUN 104 LUN 53 LUN 124 LUN 130 LUN 84 LUN 43 LUN 17 LUN 30 LUN 19 LUN 88 LUN 207 LUN 26 LUN 8 LUN 2 LUN 4 LUN LUN 1 LUN 1 LUN 1 LUN 1 LUN 1 LUN 5 LUN 6 LUN 5 LUN LUN LUN LUN LUN LUN LUN LUN LUN LUN LUN LUN LUN LU LU LU LUN LUN LUN LUN LU LU LU LU LU LU LU LU L L L L L L L L L L L L L L L L L c0vmware13p, c0vmware...c0vmware13p, c0vmware... c0file01p, c0file02p,...c0file01p, c0file02p,... c0visrptdb01pc0visrptdb01p c0baltimore01pc0baltimore01p c0ws01p, c0ws02pc0ws01p, c0ws02p c0orademand02tc0orademand02t LUN 200 LUN 205 LUN 94 LUN 81 LUN 78 LUN 82 LUN 77 LUN 35 LUN 49 LUN 104 LUN 53 LUN 124 LUN 130 LUN 84 LUN 43 LUN 17 LUN 30 LUN 19 LUN 88 LUN 207 LUN 26 LUN 8 LUN 2 LUN 4 LUN LUN 1 LUN 1 LUN 1 LUN 1 LUN 1 LUN 5 LUN 6 LUN 5 LUN LUN LUN LUN LUN LUN LUN LUN LUN LUN LUN LUN LUN LU LU LU LUN LUN LUN LUN LU LU LU LU LU LU LU LU L L L L L L L L L L L L L L L L L c0vmware13p, c0vmware... c0file01p, c0file02p,... c0visrptdb01p c0baltimore01p c0ws01p, c0ws02p c0orademand02t
- 32. Object Optimizations Oracle's default tablespace settings waste disk and create unnecessary I/O Often a smallfile tablespace is a bad choice Default table creations are almost always inefficient Often heap tables are a bad choice Not every index should be a default B*Tree And poor choices lead to poor performance Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 32 And poor choices lead to poor performance Oracle sequence object defaults are almost always inefficient A default cache size of 20 is 99.9% of the time a bad choice The vast majority of PL/SQL I see written would have been valid in version 7.3.4 I've almost never seen constants declared properly Oracle optimizes data dictionary performance by clustering tables ... do you?
- 33. I'm Not Afraid To Show You Mine Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 33
- 34. Page One Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 34
- 35. Page Two Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 35
- 36. Page Three Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 36
- 37. Page Four Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 37
- 38. Page Five Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 38
- 39. Optimizer Plans (1:4) SELECT DISTINCT E1_2.OBJECT_ID FROM PMCM.ELEMENT_DETAIL E1_1, PMCM.ELEMENT_DETAIL E1_2, PMCM.MARK_NETW_HIERARCHY H1, PMCM.ELEMENT_DETAIL E2_1, PMCM.ELEMENT_DETAIL E2_2, PMCM.MARK_NETW_HIERARCHY H2 WHERE E1_1.OBJECT_ID = H1.PARENT_ID AND E1_2.OBJECT_ID = H1.OBJECT_ID AND E2_1.OBJECT_ID = H2.PARENT_ID AND E2_2.OBJECT_ID = H2.OBJECT_ID AND E1_1.CURRENT_IND = 'Y' AND E2_1.CURRENT_IND = 'Y' AND E2_1.CURRENT_IND = 'Y' AND E2_2.CURRENT_IND = 'Y' AND H1.CURRENT_IND = 'Y' AND H2.CURRENT_IND = 'Y' AND H1.HIERARCHY_TYPE = 'NETWORK' AND H2.HIERARCHY_TYPE = 'NETWORK' AND H1.PARENT_TYPE IN ('BSC','RNC') AND H2.PARENT_TYPE IN ('BSC','RNC') AND E2_2.ELEMENT_TYPE = 'CELL' AND E1_2.ELEMENT_TYPE = 'CELL' AND H1.PARENT_TYPE IN ('BSC','RNC') AND E1_1.ELEMENT_NAME = E2_1.ELEMENT_NAME AND E1_1.ELEMENT_ID = E2_1.ELEMENT_ID AND E1_2.ELEMENT_NAME = E2_2.ELEMENT_NAME AND E1_2.ELEMENT_ID = E2_2.ELEMENT_ID AND E1_2.USEID LIKE '*%' AND E2_2.USEID NOT LIKE '*%'; ---------------------------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | Pstart| Pstop | --------------------------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 78 | | 74M (40)| 50:54:42 | | | | 1 | TEMP TABLE TRANSFORMATION | | | | | | | | | SELECT DISTINCT E1_2.OBJECT_ID FROM PMCM.ELEMENT_DETAIL E1_1, PMCM.ELEMENT_DETAIL E1_2, PMCM.MARK_NETW_HIERARCHY H1, PMCM.ELEMENT_DETAIL E2_1, PMCM.ELEMENT_DETAIL E2_2, PMCM.MARK_NETW_HIERARCHY H2 WHERE E1_1.OBJECT_ID = H1.PARENT_ID AND E1_2.OBJECT_ID = H1.OBJECT_ID AND E2_1.OBJECT_ID = H2.PARENT_ID AND E2_2.OBJECT_ID = H2.OBJECT_ID AND E1_1.CURRENT_IND = 'Y' AND E2_1.CURRENT_IND = 'Y' AND E2_1.CURRENT_IND = 'Y' AND E2_2.CURRENT_IND = 'Y' AND H1.CURRENT_IND = 'Y' AND H2.CURRENT_IND = 'Y' AND H1.HIERARCHY_TYPE = 'NETWORK' AND H2.HIERARCHY_TYPE = 'NETWORK' AND H1.PARENT_TYPE IN ('BSC','RNC') AND H2.PARENT_TYPE IN ('BSC','RNC') AND E2_2.ELEMENT_TYPE = 'CELL' AND E1_2.ELEMENT_TYPE = 'CELL' AND H1.PARENT_TYPE IN ('BSC','RNC') AND E1_1.ELEMENT_NAME = E2_1.ELEMENT_NAME AND E1_1.ELEMENT_ID = E2_1.ELEMENT_ID AND E1_2.ELEMENT_NAME = E2_2.ELEMENT_NAME AND E1_2.ELEMENT_ID = E2_2.ELEMENT_ID AND E1_2.USEID LIKE '*%' AND E2_2.USEID NOT LIKE '*%'; ---------------------------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | Pstart| Pstop | --------------------------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 78 | | 74M (40)| 50:54:42 | | | | 1 | TEMP TABLE TRANSFORMATION | | | | | | | | | Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 39 | 1 | TEMP TABLE TRANSFORMATION | | | | | | | | | | 2 | LOAD AS SELECT | | | | | | | | | | 3 | PARTITION RANGE ALL | | 22M| 1111M| | 38153 (11)| 00:01:34 | 1 | 29 | |* 4 | TABLE ACCESS FULL | ELEMENT_DETAIL | 22M| 1111M| | 38153 (11)| 00:01:34 | | | | 5 | LOAD AS SELECT | | | | | | | | | | 6 | PARTITION HASH ALL | | 337K| 9231K| | 3514 (15)| 00:00:09 | 1 | 16 | |* 7 | TABLE ACCESS FULL | MARK_NETW_HIERARCHY | 337K| 9231K| | 3514 (15)| 00:00:09 | | | | 8 | SORT AGGREGATE | | 1 | 78 | | | | | | |* 9 | HASH JOIN | | 927G| 65T| 534M| 74M (40)| 50:53:00 | | | | 10 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 11 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485F_6A66C42E | 22M| 1111M| | 16808 (12)| 00 | | | |* 12 | HASH JOIN | | 21G| 1272G| 534M| 1616K (43)| 01:06:04 | | | | 13 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 14 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485F_6A66C42E | 22M| 1111M| | 16808 (12)| 0 | | | |* 15 | HASH JOIN | | 476M| 23G| 524M| 97327 (22)| 00:03:59 | | | |* 16 | HASH JOIN | | 10M| 401M| 8704K| 34520 (10)| 00:01:25 | | | |* 17 | HASH JOIN | | 234K| 5948K| 8256K| 783 (10)| 00:00:02 | | | | 18 | VIEW | | 337K| 4286K| | 142 (14)| 00:00:01 | | | | 19 | TABLE ACCESS FULL | SYS_TEMP_0FDA74860_6A66C42E | 337K| 3956K| | 142 (14)| 00:00:01 | | | | 20 | VIEW | | 337K| 4286K| | 142 (14)| 00:00:01 | | | | 21 | TABLE ACCESS FULL | SYS_TEMP_0FDA74860_6A66C42E | 337K| 3956K| | 142 (14)| 00:00:01 | | | | 22 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 23 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485F_6A66C42E | 22M| 1111M| | 16808 (12)| 00:00:42 | | | | 24 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 25 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485F_6A66C42E | 22M| 1111M| | 16808 (12)| 0 | | | --------------------------------------------------------------------------------------------------------------------------------- | 1 | TEMP TABLE TRANSFORMATION | | | | | | | | | | 2 | LOAD AS SELECT | | | | | | | | | | 3 | PARTITION RANGE ALL | | 22M| 1111M| | 38153 (11)| 00:01:34 | 1 | 29 | |* 4 | TABLE ACCESS FULL | ELEMENT_DETAIL | 22M| 1111M| | 38153 (11)| 00:01:34 | | | | 5 | LOAD AS SELECT | | | | | | | | | | 6 | PARTITION HASH ALL | | 337K| 9231K| | 3514 (15)| 00:00:09 | 1 | 16 | |* 7 | TABLE ACCESS FULL | MARK_NETW_HIERARCHY | 337K| 9231K| | 3514 (15)| 00:00:09 | | | | 8 | SORT AGGREGATE | | 1 | 78 | | | | | | |* 9 | HASH JOIN | | 927G| 65T| 534M| 74M (40)| 50:53:00 | | | | 10 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 11 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485F_6A66C42E | 22M| 1111M| | 16808 (12)| 00 | | | |* 12 | HASH JOIN | | 21G| 1272G| 534M| 1616K (43)| 01:06:04 | | | | 13 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 14 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485F_6A66C42E | 22M| 1111M| | 16808 (12)| 0 | | | |* 15 | HASH JOIN | | 476M| 23G| 524M| 97327 (22)| 00:03:59 | | | |* 16 | HASH JOIN | | 10M| 401M| 8704K| 34520 (10)| 00:01:25 | | | |* 17 | HASH JOIN | | 234K| 5948K| 8256K| 783 (10)| 00:00:02 | | | | 18 | VIEW | | 337K| 4286K| | 142 (14)| 00:00:01 | | | | 19 | TABLE ACCESS FULL | SYS_TEMP_0FDA74860_6A66C42E | 337K| 3956K| | 142 (14)| 00:00:01 | | | | 20 | VIEW | | 337K| 4286K| | 142 (14)| 00:00:01 | | | | 21 | TABLE ACCESS FULL | SYS_TEMP_0FDA74860_6A66C42E | 337K| 3956K| | 142 (14)| 00:00:01 | | | | 22 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 23 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485F_6A66C42E | 22M| 1111M| | 16808 (12)| 00:00:42 | | | | 24 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 25 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485F_6A66C42E | 22M| 1111M| | 16808 (12)| 0 | | | ---------------------------------------------------------------------------------------------------------------------------------
- 40. Optimizer Plans (2:4) ---------------------------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | Pstart| Pstop | ---------------------------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 78 | | 14T(100)|999:59:59 | | | | 1 | TEMP TABLE TRANSFORMATION | | | | | | | | | | 2 | LOAD AS SELECT | | | | | | | | | | 3 | PARTITION RANGE ALL | | 22M| 1111M| | 38153 (11)| 00:01:34 | 1 | 29 | |* 4 | TABLE ACCESS FULL | ELEMENT_DETAIL | 22M| 1111M| | 38153 (11)| 00:01:34 | | | | 5 | LOAD AS SELECT | | | | | | | | | | 6 | PARTITION HASH ALL | | 337K| 9231K| | 3514 (15)| 00:00:09 | 1 | 16 | |* 7 | TABLE ACCESS FULL | MARK_NETW_HIERARCHY | 337K| 9231K| | 3514 (15)| 00:00:09 | | | | 8 | SORT AGGREGATE | | 1 | 78 | | | | | | | 9 | MERGE JOIN | | 471P| 15E| | 14T(100)|999:59:59 | | | | 10 | MERGE JOIN | | 10P| 616P| | 694G (81)|999:59:59 | | | | 11 | MERGE JOIN | | 231T| 10P| | 377G (64)|999:59:59 | | | | 12 | SORT JOIN | | 334T| 11P| 28P| 377G (64)|999:59:59 | | | | 13 | MERGE JOIN CARTESIAN| | 334T| 11P| | 140G (14)|999:59:59 | | | |* 14 | HASH JOIN | | 989M| 23G| 534M| 96010 (38)| 00:03:56 | | | | 15 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 16 | TABLE ACCESS FULL| SYS_TEMP_0FDA7485B_6A66C42E | 22M| 1111M| | 16808 (12)| 00:00:42 | | | | 17 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 18 | TABLE ACCESS FULL| SYS_TEMP_0FDA7485B_6A66C42E | 22M| 1111M| | 16808 (12)| 00:00:42 | | | | 19 | BUFFER SORT | | 337K| 4286K| | 140G (14)|999:59:59 | | | | 20 | VIEW | | 337K| 4286K| | 142 (14)| 00:00:01 | | | | 21 | TABLE ACCESS FULL| SYS_TEMP_0FDA7485C_6A66C42E | 337K| 3956K| | 142 (14)| 00:00:01 | | | ---------------------------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | Pstart| Pstop | ---------------------------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 78 | | 14T(100)|999:59:59 | | | | 1 | TEMP TABLE TRANSFORMATION | | | | | | | | | | 2 | LOAD AS SELECT | | | | | | | | | | 3 | PARTITION RANGE ALL | | 22M| 1111M| | 38153 (11)| 00:01:34 | 1 | 29 | |* 4 | TABLE ACCESS FULL | ELEMENT_DETAIL | 22M| 1111M| | 38153 (11)| 00:01:34 | | | | 5 | LOAD AS SELECT | | | | | | | | | | 6 | PARTITION HASH ALL | | 337K| 9231K| | 3514 (15)| 00:00:09 | 1 | 16 | |* 7 | TABLE ACCESS FULL | MARK_NETW_HIERARCHY | 337K| 9231K| | 3514 (15)| 00:00:09 | | | | 8 | SORT AGGREGATE | | 1 | 78 | | | | | | | 9 | MERGE JOIN | | 471P| 15E| | 14T(100)|999:59:59 | | | | 10 | MERGE JOIN | | 10P| 616P| | 694G (81)|999:59:59 | | | | 11 | MERGE JOIN | | 231T| 10P| | 377G (64)|999:59:59 | | | | 12 | SORT JOIN | | 334T| 11P| 28P| 377G (64)|999:59:59 | | | | 13 | MERGE JOIN CARTESIAN| | 334T| 11P| | 140G (14)|999:59:59 | | | |* 14 | HASH JOIN | | 989M| 23G| 534M| 96010 (38)| 00:03:56 | | | | 15 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 16 | TABLE ACCESS FULL| SYS_TEMP_0FDA7485B_6A66C42E | 22M| 1111M| | 16808 (12)| 00:00:42 | | | | 17 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 18 | TABLE ACCESS FULL| SYS_TEMP_0FDA7485B_6A66C42E | 22M| 1111M| | 16808 (12)| 00:00:42 | | | | 19 | BUFFER SORT | | 337K| 4286K| | 140G (14)|999:59:59 | | | | 20 | VIEW | | 337K| 4286K| | 142 (14)| 00:00:01 | | | | 21 | TABLE ACCESS FULL| SYS_TEMP_0FDA7485C_6A66C42E | 337K| 3956K| | 142 (14)| 00:00:01 | | | Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 40 | 21 | TABLE ACCESS FULL| SYS_TEMP_0FDA7485C_6A66C42E | 337K| 3956K| | 142 (14)| 00:00:01 | | | |* 22 | SORT JOIN | | 337K| 4286K| 12M| 844 (14)| 00:00:03 | | | | 23 | VIEW | | 337K| 4286K| | 142 (14)| 00:00:01 | | | | 24 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485C_6A66C42E | 337K| 3956K| | 142 (14)| 00:00:01 | | | |* 25 | SORT JOIN | | 22M| 277M| 855M| 65084 (16)| 00:02:40 | | | | 26 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 27 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485B_6A66C42E | 22M| 1111M| | 16808 (12)| 0 | | | |* 28 | SORT JOIN | | 22M| 277M| 855M| 65084 (16)| 00:02:40 | | | | 29 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 30 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485B_6A66C42E | 22M| 1111M| | 16808 (12)| 0 | | | ---------------------------------------------------------------------------------------------------------------------------------- | 21 | TABLE ACCESS FULL| SYS_TEMP_0FDA7485C_6A66C42E | 337K| 3956K| | 142 (14)| 00:00:01 | | | |* 22 | SORT JOIN | | 337K| 4286K| 12M| 844 (14)| 00:00:03 | | | | 23 | VIEW | | 337K| 4286K| | 142 (14)| 00:00:01 | | | | 24 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485C_6A66C42E | 337K| 3956K| | 142 (14)| 00:00:01 | | | |* 25 | SORT JOIN | | 22M| 277M| 855M| 65084 (16)| 00:02:40 | | | | 26 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 27 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485B_6A66C42E | 22M| 1111M| | 16808 (12)| 0 | | | |* 28 | SORT JOIN | | 22M| 277M| 855M| 65084 (16)| 00:02:40 | | | | 29 | VIEW | | 22M| 277M| | 16808 (12)| 00:00:42 | | | | 30 | TABLE ACCESS FULL | SYS_TEMP_0FDA7485B_6A66C42E | 22M| 1111M| | 16808 (12)| 0 | | | ----------------------------------------------------------------------------------------------------------------------------------
- 41. Optimizer Plans (3:4) WITH ed AS (SELECT object_id, element_id, element_name, element_type, useid FROM pmcm.element_detail WHERE element_type = 'CELL' AND current_ind = 'Y'), mnh AS (SELECT parent_id, object_id FROM pmcm.mark_netw_hierarchy WHERE current_ind = 'Y' AND hierarchy_type = 'NETWORK' AND parent_type IN ('BSC', 'RNC')) SELECT COUNT(*) FROM ed e1_1, ed e1_2, ed e2_1, ed e2_2, mnh h1, mnh h2 WHERE e1_1.object_id = h1.parent_id AND e1_2.object_id = h1.object_id AND e2_1.object_id = h2.parent_id AND e2_2.object_id = h2.object_id AND e1_1.element_name = e2_1.element_name AND e1_1.element_id = e2_1.element_id AND e1_2.element_name = e2_2.element_name AND e1_2.element_id = e2_2.element_id AND e1_2.useid LIKE '*%' AND e2_2.useid NOT LIKE '*%'; ----------------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | ------------------------------------------------------------------------------------------------------------------------ | 0 | SELECT STATEMENT | | 1 | 214 | | 100K (6)| 00:04:08 | | 1 | HASH UNIQUE | | 1 | 214 | | 100K (6)| 00:04:08 | WITH ed AS (SELECT object_id, element_id, element_name, element_type, useid FROM pmcm.element_detail WHERE element_type = 'CELL' AND current_ind = 'Y'), mnh AS (SELECT parent_id, object_id FROM pmcm.mark_netw_hierarchy WHERE current_ind = 'Y' AND hierarchy_type = 'NETWORK' AND parent_type IN ('BSC', 'RNC')) SELECT COUNT(*) FROM ed e1_1, ed e1_2, ed e2_1, ed e2_2, mnh h1, mnh h2 WHERE e1_1.object_id = h1.parent_id AND e1_2.object_id = h1.object_id AND e2_1.object_id = h2.parent_id AND e2_2.object_id = h2.object_id AND e1_1.element_name = e2_1.element_name AND e1_1.element_id = e2_1.element_id AND e1_2.element_name = e2_2.element_name AND e1_2.element_id = e2_2.element_id AND e1_2.useid LIKE '*%' AND e2_2.useid NOT LIKE '*%'; ----------------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | ------------------------------------------------------------------------------------------------------------------------ | 0 | SELECT STATEMENT | | 1 | 214 | | 100K (6)| 00:04:08 | | 1 | HASH UNIQUE | | 1 | 214 | | 100K (6)| 00:04:08 | Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 41 | 1 | HASH UNIQUE | | 1 | 214 | | 100K (6)| 00:04:08 | |* 2 | HASH JOIN | | 1 | 214 | 12M| 100K (6)| 00:04:08 | | 3 | PARTITION HASH ALL | | 337K| 9231K| | 3514 (15)| 00:00:09 | |* 4 | TABLE ACCESS FULL | MARK_NETW_HIERARCHY | 337K| 9231K| | 3514 (15)| 00:00:00 | |* 5 | HASH JOIN | | 207K| 36M| 22M| 95860 (6)| 00:03:56 | | 6 | PARTITION RANGE ALL | | 586K| 15M| | 16233 (2)| 00:00:40 | | 7 | TABLE ACCESS BY LOCAL INDEX ROWID | ELEMENT_DETAIL | 586K| 15M| | 16233 | ??:??:?? | |* 8 | INDEX SKIP SCAN | ED_ET_TECH_CI | 586K| | | 12791 (1)| 00:00:3? | |* 9 | HASH JOIN | | 207K| 31M| 22M| 77982 (7)| 00:03:12 | | 10 | PARTITION RANGE ALL | | 586K| 15M| | 16233 (2)| 00:00:40 | | 11 | TABLE ACCESS BY LOCAL INDEX ROWID | ELEMENT_DETAIL | 586K| 15M| | 16233 | ??:??:?? | |* 12 | INDEX SKIP SCAN | ED_ET_TECH_CI | 586K| | | 12791 (1)| 00:00:?? | |* 13 | HASH JOIN | | 179K| 22M| 12M| 60372 (8)| 00:02:29 | | 14 | PARTITION HASH ALL | | 337K| 9231K| | 3514 (15)| 00:00:09 | |* 15 | TABLE ACCESS FULL | MARK_NETW_HIERARCHY | 337K| 9231K| | 3514 (15)| 00:00:?? | |* 16 | HASH JOIN | | 184K| 17M| 10M| 55886 (8)| 00:02:18 | | 17 | PARTITION RANGE ALL | | 184K| 9008K| | 37137 (8)| 00:01:32 | |* 18 | TABLE ACCESS FULL | ELEMENT_DETAIL | 184K| 9008K| | 37137 (8)| 00:01:32 | | 19 | PARTITION RANGE ALL | | 576K| 28M| | 17383 (8)| 00:00:43 | |* 20 | TABLE ACCESS BY LOCAL INDEX ROWID| ELEMENT_DETAIL | 576K| 28M| | 17383 (8)| ??:??:?? | |* 21 | INDEX SKIP SCAN | ED_ET_TECH_CI | 583K| | | 13939 (9)| 00:00:35 | ------------------------------------------------------------------------------------------------------------------------ | 1 | HASH UNIQUE | | 1 | 214 | | 100K (6)| 00:04:08 | |* 2 | HASH JOIN | | 1 | 214 | 12M| 100K (6)| 00:04:08 | | 3 | PARTITION HASH ALL | | 337K| 9231K| | 3514 (15)| 00:00:09 | |* 4 | TABLE ACCESS FULL | MARK_NETW_HIERARCHY | 337K| 9231K| | 3514 (15)| 00:00:00 | |* 5 | HASH JOIN | | 207K| 36M| 22M| 95860 (6)| 00:03:56 | | 6 | PARTITION RANGE ALL | | 586K| 15M| | 16233 (2)| 00:00:40 | | 7 | TABLE ACCESS BY LOCAL INDEX ROWID | ELEMENT_DETAIL | 586K| 15M| | 16233 | ??:??:?? | |* 8 | INDEX SKIP SCAN | ED_ET_TECH_CI | 586K| | | 12791 (1)| 00:00:3? | |* 9 | HASH JOIN | | 207K| 31M| 22M| 77982 (7)| 00:03:12 | | 10 | PARTITION RANGE ALL | | 586K| 15M| | 16233 (2)| 00:00:40 | | 11 | TABLE ACCESS BY LOCAL INDEX ROWID | ELEMENT_DETAIL | 586K| 15M| | 16233 | ??:??:?? | |* 12 | INDEX SKIP SCAN | ED_ET_TECH_CI | 586K| | | 12791 (1)| 00:00:?? | |* 13 | HASH JOIN | | 179K| 22M| 12M| 60372 (8)| 00:02:29 | | 14 | PARTITION HASH ALL | | 337K| 9231K| | 3514 (15)| 00:00:09 | |* 15 | TABLE ACCESS FULL | MARK_NETW_HIERARCHY | 337K| 9231K| | 3514 (15)| 00:00:?? | |* 16 | HASH JOIN | | 184K| 17M| 10M| 55886 (8)| 00:02:18 | | 17 | PARTITION RANGE ALL | | 184K| 9008K| | 37137 (8)| 00:01:32 | |* 18 | TABLE ACCESS FULL | ELEMENT_DETAIL | 184K| 9008K| | 37137 (8)| 00:01:32 | | 19 | PARTITION RANGE ALL | | 576K| 28M| | 17383 (8)| 00:00:43 | |* 20 | TABLE ACCESS BY LOCAL INDEX ROWID| ELEMENT_DETAIL | 576K| 28M| | 17383 (8)| ??:??:?? | |* 21 | INDEX SKIP SCAN | ED_ET_TECH_CI | 583K| | | 13939 (9)| 00:00:35 | ------------------------------------------------------------------------------------------------------------------------
- 42. Optimizer Plans (4:4) --------------------------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | --------------------------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | | | | 264T(100)| | |* 1 | VIEW | | 156P| 15E| | 264T (79)|999:59:59 | |* 2 | WINDOW SORT PUSHED RANK | | 156P| 15E| 15E| 264T (79)|999:59:59 | | 3 | MERGE JOIN CARTESIAN | | 156P| 15E| | 68T (16)|999:59:59 | | 4 | MERGE JOIN CARTESIAN | | 220G| 205T| | 96M (16)| 26:57:48 | | 5 | MERGE JOIN CARTESIAN | | 310K| 302M| | 232 (11)| 00:00:01 | | 6 | MERGE JOIN CARTESIAN | | 779 | 777K| | 22 (0)| 00:00:01 | | 7 | NESTED LOOPS | | | | | | | | 8 | NESTED LOOPS | | 2 | 2044 | | 20 (0)| 00:00:01 | | 9 | NESTED LOOPS OUTER | | 2 | 1990 | | 18 (0)| 00:00:01 | | 10 | NESTED LOOPS | | 2 | 1868 | | 17 (0)| 00:00:01 | | 11 | NESTED LOOPS | | 2 | 1712 | | 15 (0)| 00:00:01 | | 12 | NESTED LOOPS | | 2 | 1564 | | 13 (0)| 00:00:01 | | 13 | MERGE JOIN CARTESIAN | | 2 | 1442 | | 11 (0)| 00:00:01 | | 14 | NESTED LOOPS OUTER | | 1 | 625 | | 8 (0)| 00:00:01 | | 15 | NESTED LOOPS OUTER | | 1 | 613 | | 7 (0)| 00:00:01 | | 16 | NESTED LOOPS | | 1 | 580 | | 6 (0)| 00:00:01 | | 17 | NESTED LOOPS OUTER | | 1 | 539 | | 5 (0)| 00:00:01 | | 18 | NESTED LOOPS OUTER | | 1 | 340 | | 5 (0)| 00:00:01 | | 19 | TABLE ACCESS BY INDEX ROWID| PA_STUDENT | 1 | 316 | | 3 (0)| 00:00:01 | |* 20 | INDEX UNIQUE SCAN | PK_STUDENT | 1 | | | 2 (0)| 00:00:01 | | 21 | TABLE ACCESS BY INDEX ROWID| PA_STUD_USER | 1 | 24 | | 2 (0)| 00:00:01 | |* 22 | INDEX UNIQUE SCAN | PK_STUD_USER | 1 | | | 1 (0)| 00:00:01 | | 23 | TABLE ACCESS BY INDEX ROWID | PA_ORG | 1 | 199 | | 0 (0)| | |* 24 | INDEX UNIQUE SCAN | PK_ORG | 1 | | | 0 (0)| | | 25 | TABLE ACCESS BY INDEX ROWID | PA_DOMAIN | 13 | 533 | | 1 (0)| 00:00:01 | --------------------------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | --------------------------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | | | | 264T(100)| | |* 1 | VIEW | | 156P| 15E| | 264T (79)|999:59:59 | |* 2 | WINDOW SORT PUSHED RANK | | 156P| 15E| 15E| 264T (79)|999:59:59 | | 3 | MERGE JOIN CARTESIAN | | 156P| 15E| | 68T (16)|999:59:59 | | 4 | MERGE JOIN CARTESIAN | | 220G| 205T| | 96M (16)| 26:57:48 | | 5 | MERGE JOIN CARTESIAN | | 310K| 302M| | 232 (11)| 00:00:01 | | 6 | MERGE JOIN CARTESIAN | | 779 | 777K| | 22 (0)| 00:00:01 | | 7 | NESTED LOOPS | | | | | | | | 8 | NESTED LOOPS | | 2 | 2044 | | 20 (0)| 00:00:01 | | 9 | NESTED LOOPS OUTER | | 2 | 1990 | | 18 (0)| 00:00:01 | | 10 | NESTED LOOPS | | 2 | 1868 | | 17 (0)| 00:00:01 | | 11 | NESTED LOOPS | | 2 | 1712 | | 15 (0)| 00:00:01 | | 12 | NESTED LOOPS | | 2 | 1564 | | 13 (0)| 00:00:01 | | 13 | MERGE JOIN CARTESIAN | | 2 | 1442 | | 11 (0)| 00:00:01 | | 14 | NESTED LOOPS OUTER | | 1 | 625 | | 8 (0)| 00:00:01 | | 15 | NESTED LOOPS OUTER | | 1 | 613 | | 7 (0)| 00:00:01 | | 16 | NESTED LOOPS | | 1 | 580 | | 6 (0)| 00:00:01 | | 17 | NESTED LOOPS OUTER | | 1 | 539 | | 5 (0)| 00:00:01 | | 18 | NESTED LOOPS OUTER | | 1 | 340 | | 5 (0)| 00:00:01 | | 19 | TABLE ACCESS BY INDEX ROWID| PA_STUDENT | 1 | 316 | | 3 (0)| 00:00:01 | |* 20 | INDEX UNIQUE SCAN | PK_STUDENT | 1 | | | 2 (0)| 00:00:01 | | 21 | TABLE ACCESS BY INDEX ROWID| PA_STUD_USER | 1 | 24 | | 2 (0)| 00:00:01 | |* 22 | INDEX UNIQUE SCAN | PK_STUD_USER | 1 | | | 1 (0)| 00:00:01 | | 23 | TABLE ACCESS BY INDEX ROWID | PA_ORG | 1 | 199 | | 0 (0)| | |* 24 | INDEX UNIQUE SCAN | PK_ORG | 1 | | | 0 (0)| | | 25 | TABLE ACCESS BY INDEX ROWID | PA_DOMAIN | 13 | 533 | | 1 (0)| 00:00:01 | Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 42 | 25 | TABLE ACCESS BY INDEX ROWID | PA_DOMAIN | 13 | 533 | | 1 (0)| 00:00:01 | |* 26 | INDEX UNIQUE SCAN | PK_DOMAIN | 1 | | | 0 (0)| | | 27 | TABLE ACCESS BY INDEX ROWID | PA_USRRF_STUD | 100 | 3300 | | 1 (0)| 00:00:01 | |* 28 | INDEX UNIQUE SCAN | PK_USRRF_STUD | 1 | | | 0 (0)| | | 29 | VIEW PUSHED PREDICATE | PV_STUD_USER | 1 | 12 | | 1 (0)| 00:00:01 | |* 30 | FILTER | | | | | | | | 31 | NESTED LOOPS OUTER | | 1 | 22 | | 264 (11)| 00:00:01 | |* 32 | INDEX UNIQUE SCAN | PK_STUDENT | 1 | 10 | | 2 (0)| 00:00:01 | |* 33 | MAT_VIEW ACCESS FULL | PV_AP_STUD_USER | 1 | 12 | | 262 (11)| 00:00:01 | | 34 | BUFFER SORT | | 2 | 192 | | 10 (0)| 00:00:01 | | 35 | TABLE ACCESS BY INDEX ROWID | PA_CPNT_COMPLIANCE_DATA | 2 | 192 | | 3 (0)| 00:00:01 | |* 36 | INDEX RANGE SCAN | IX_CPNT_CD__EVTHST | 2 | | | 1 (0)| 00:00:01 | | 37 | TABLE ACCESS BY INDEX ROWID | PA_CPNT_TYPE | 1 | 61 | | 1 (0)| 00:00:01 | |* 38 | INDEX UNIQUE SCAN | PK_CPNT_TYPE | 1 | | | 0 (0)| | | 39 | TABLE ACCESS BY INDEX ROWID | PA_RQMT_TYPE | 1 | 74 | | 1 (0)| 00:00:01 | |* 40 | INDEX UNIQUE SCAN | PK_RQMT_TYPE | 1 | | | 0 (0)| | | 41 | TABLE ACCESS BY INDEX ROWID | PA_CMPL_STAT | 1 | 78 | | 1 (0)| 00:00:01 | |* 42 | INDEX UNIQUE SCAN | PK_CMPL_STAT | 1 | | | 0 (0)| | | 43 | TABLE ACCESS BY INDEX ROWID | PA_QUAL | 1 | 61 | | 1 (0)| 00:00:01 | |* 44 | INDEX UNIQUE SCAN | PK_QUAL | 1 | | | 0 (0)| | |* 45 | INDEX UNIQUE SCAN | PK_CPNT | 1 | | | 0 (0)| | | 46 | TABLE ACCESS BY INDEX ROWID | PA_CPNT | 1 | 27 | | 1 (0)| 00:00:01 | | 47 | BUFFER SORT | | 399 | | | 21 (0)| 00:00:01 | | 48 | INDEX FAST FULL SCAN | PK_USRRF_STUD | 399 | | | 1 (0)| 00:00:01 | | 49 | BUFFER SORT | | 399 | | | 231 (11)| 00:00:01 | | 50 | INDEX FAST FULL SCAN | PK_USRRF_STUD | 399 | | | 0 (0)| | | 51 | BUFFER SORT | | 710K| | | 96M (16)| 26:57:48 | | 52 | INDEX FAST FULL SCAN | IX_STUD_USER__STUDENT | 710K| | | 309 (16)| 00:00:01 | | 53 | BUFFER SORT | | 710K| | | 264T (79)|999:59:59 | | 54 | INDEX FAST FULL SCAN | IX_STUD_USER__STUDENT | 710K| | | 309 (16)| 00:00:01 | --------------------------------------------------------------------------------------------------------------------------------- | 25 | TABLE ACCESS BY INDEX ROWID | PA_DOMAIN | 13 | 533 | | 1 (0)| 00:00:01 | |* 26 | INDEX UNIQUE SCAN | PK_DOMAIN | 1 | | | 0 (0)| | | 27 | TABLE ACCESS BY INDEX ROWID | PA_USRRF_STUD | 100 | 3300 | | 1 (0)| 00:00:01 | |* 28 | INDEX UNIQUE SCAN | PK_USRRF_STUD | 1 | | | 0 (0)| | | 29 | VIEW PUSHED PREDICATE | PV_STUD_USER | 1 | 12 | | 1 (0)| 00:00:01 | |* 30 | FILTER | | | | | | | | 31 | NESTED LOOPS OUTER | | 1 | 22 | | 264 (11)| 00:00:01 | |* 32 | INDEX UNIQUE SCAN | PK_STUDENT | 1 | 10 | | 2 (0)| 00:00:01 | |* 33 | MAT_VIEW ACCESS FULL | PV_AP_STUD_USER | 1 | 12 | | 262 (11)| 00:00:01 | | 34 | BUFFER SORT | | 2 | 192 | | 10 (0)| 00:00:01 | | 35 | TABLE ACCESS BY INDEX ROWID | PA_CPNT_COMPLIANCE_DATA | 2 | 192 | | 3 (0)| 00:00:01 | |* 36 | INDEX RANGE SCAN | IX_CPNT_CD__EVTHST | 2 | | | 1 (0)| 00:00:01 | | 37 | TABLE ACCESS BY INDEX ROWID | PA_CPNT_TYPE | 1 | 61 | | 1 (0)| 00:00:01 | |* 38 | INDEX UNIQUE SCAN | PK_CPNT_TYPE | 1 | | | 0 (0)| | | 39 | TABLE ACCESS BY INDEX ROWID | PA_RQMT_TYPE | 1 | 74 | | 1 (0)| 00:00:01 | |* 40 | INDEX UNIQUE SCAN | PK_RQMT_TYPE | 1 | | | 0 (0)| | | 41 | TABLE ACCESS BY INDEX ROWID | PA_CMPL_STAT | 1 | 78 | | 1 (0)| 00:00:01 | |* 42 | INDEX UNIQUE SCAN | PK_CMPL_STAT | 1 | | | 0 (0)| | | 43 | TABLE ACCESS BY INDEX ROWID | PA_QUAL | 1 | 61 | | 1 (0)| 00:00:01 | |* 44 | INDEX UNIQUE SCAN | PK_QUAL | 1 | | | 0 (0)| | |* 45 | INDEX UNIQUE SCAN | PK_CPNT | 1 | | | 0 (0)| | | 46 | TABLE ACCESS BY INDEX ROWID | PA_CPNT | 1 | 27 | | 1 (0)| 00:00:01 | | 47 | BUFFER SORT | | 399 | | | 21 (0)| 00:00:01 | | 48 | INDEX FAST FULL SCAN | PK_USRRF_STUD | 399 | | | 1 (0)| 00:00:01 | | 49 | BUFFER SORT | | 399 | | | 231 (11)| 00:00:01 | | 50 | INDEX FAST FULL SCAN | PK_USRRF_STUD | 399 | | | 0 (0)| | | 51 | BUFFER SORT | | 710K| | | 96M (16)| 26:57:48 | | 52 | INDEX FAST FULL SCAN | IX_STUD_USER__STUDENT | 710K| | | 309 (16)| 00:00:01 | | 53 | BUFFER SORT | | 710K| | | 264T (79)|999:59:59 | | 54 | INDEX FAST FULL SCAN | IX_STUD_USER__STUDENT | 710K| | | 309 (16)| 00:00:01 | ---------------------------------------------------------------------------------------------------------------------------------
- 43. Poorly Written Applications SELECT /*+ RESULT_CACHE */ srvr_idSELECT /*+ RESULT_CACHE */ srvr_id Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 43 SELECT /*+ RESULT_CACHE */ srvr_id FROM ( SELECT srvr_id, SUM(cnt) SUMCNT FROM ( SELECT DISTINCT srvr_id, 1 AS CNT FROM servers UNION ALL SELECT DISTINCT srvr_id, 1 FROM serv_inst) GROUP BY srvr_id) WHERE sumcnt = 2; SELECT /*+ RESULT_CACHE */ srvr_id FROM ( SELECT srvr_id, SUM(cnt) SUMCNT FROM ( SELECT DISTINCT srvr_id, 1 AS CNT FROM servers UNION ALL SELECT DISTINCT srvr_id, 1 FROM serv_inst) GROUP BY srvr_id) WHERE sumcnt = 2;
- 44. Default Oracle install favors data warehouse not OLTP Optimizer Settings NAME TYPE VALUE ------------------------------------ ----------- ----------- optimizer_adaptive_features boolean TRUE optimizer_adaptive_reporting_only boolean FALSE optimizer_capture_sql_plan_baselines boolean FALSE optimizer_dynamic_sampling integer 2 optimizer_features_enable string 12.1.0.2 optimizer_index_caching integer 0 optimizer_index_cost_adj integer 100 optimizer_mode string ALL_ROWS optimizer_secure_view_merging boolean TRUE optimizer_use_invisible_indexes boolean FALSE optimizer_use_pending_statistics boolean FALSE optimizer_use_sql_plan_baselines boolean TRUE Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 44 SQL> show parameter optimizer_mode NAME TYPE VALUE ------------------------------------ ----------- --------- optimizer_mode string ALL_ROWS SQL> ALTER SYSTEM SET OPTIMIZER_MODE='FIRST_ROWS_10'; System altered. SQL> show parameter optimizer_mode NAME TYPE VALUE ------------------------ ----------- -------------- optimizer_mode string FIRST_ROWS_10 SQL> show parameter optimizer_mode NAME TYPE VALUE ------------------------------------ ----------- --------- optimizer_mode string ALL_ROWS SQL> ALTER SYSTEM SET OPTIMIZER_MODE='FIRST_ROWS_10'; System altered. SQL> show parameter optimizer_mode NAME TYPE VALUE ------------------------ ----------- -------------- optimizer_mode string FIRST_ROWS_10 Best for Data Warehouse Best for OLTP
- 45. Optimizer Statistics Gather System Stats Gather Fixed Object Stats Gather Dictionary Stats Gather Table Stats Gather Column Stats Gather Index Stats Gather Extended Stats Stat Generation SQL> select type, count(*) 2 from v$fixed_table 3 group by type 4 order by 1; TYPE COUNT(*) ----- ---------- Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 45 Stat Generation Copy Stats Set Stats Fixing Stats ----- ---------- TABLE 1144 VIEW 1261 SQL> select name 2 from v$fixed_table 3 where rownum < 11; NAME -------------------------- X$KQFTA X$KQFVI X$KQFVT X$KQFDT X$KQFCO X$KQFOPT X$KYWMPCTAB X$KYWMWRCTAB X$KYWMCLTAB X$KYWMNF
- 46. Gathering Optimizer Statistics exec dbms_stats.set_table_prefs(USER, 'SERVERS', 'ESTIMATE_PERCENT','90'); exec dbms_stats.set_table_prefs(USER, 'SERVERS', 'DEGREE','8'); exec dbms_stats.set_table_prefs(USER, 'SERVERS', 'ESTIMATE_PERCENT','90'); exec dbms_stats.set_table_prefs(USER, 'SERVERS', 'DEGREE','8'); exec dbms_stats.gather_table_stats(USER, 'SERVERS', 'P2010103106', estimate_percent => dbms_stats.auto_sample_size, block_sample=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS', degree=>2, cascade=>TRUE exec dbms_stats.gather_table_stats(USER, 'SERVERS', 'P2010103106', estimate_percent => dbms_stats.auto_sample_size, block_sample=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS', degree=>2, cascade=>TRUE Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 46 SELECT dbms_stats.create_extended_stats(USER, 'SERV_INST', '(srvr_id, si_status)') FROM dual; SELECT dbms_stats.create_extended_stats(USER, 'SERV_INST', '(srvr_id, si_status)') FROM dual; cascade=>TRUE force=>TRUE); cascade=>TRUE force=>TRUE);
- 47. Manufacturing Optimizer Statistics exec dbms_stats.set_table_stats(USER, 'EMP', numrows=>1000000, numblks=>10000, avgrlen=>74); exec dbms_stats.set_index_stats(USER, 'ix_emp_deptno', numrows=>1000000, numlblks=>1000, numdist=>10000, clstfct=>1); exec dbms_stats.set_column_stats(USER, 'emp', 'deptno', distcnt=>10000); exec dbms_stats.set_table_stats(USER, 'dept', numrows=>100, numblks=>100); exec dbms_stats.set_table_stats(USER, 'EMP', numrows=>1000000, numblks=>10000, avgrlen=>74); exec dbms_stats.set_index_stats(USER, 'ix_emp_deptno', numrows=>1000000, numlblks=>1000, numdist=>10000, clstfct=>1); exec dbms_stats.set_column_stats(USER, 'emp', 'deptno', distcnt=>10000); exec dbms_stats.set_table_stats(USER, 'dept', numrows=>100, numblks=>100); Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 47
- 48. Root Cause Analysis by Sophisticated Guessing Check data dictionary for collected stats Explain Plans Very few people can read them ... I will prove it next AWR Reports Data <> Information If you want value from AWR ... create AWR Difference reports with DBMS_WORKLOAD_REPOSITORY.AWR_DIFF_REPORT_HTML Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 48 DBMS_WORKLOAD_REPOSITORY.AWR_DIFF_REPORT_HTML ASH Reports SQL Tuning Advisor SQL Trace and TKPROF Baselines and Evolving Baselines DBMS_TRACE, DBMS_MONITOR, TRCSESS, TRACE ANALYZER, SQLTXPLAIN
- 49. Very few people can read an explain plan Here are two plans from identical tables with identical stats Which one would you rely on? Reading Explain Plans SELECT srvr_id FROM servers INTERSECT SELECT srvr_id FROM serv_inst; SELECT srvr_id FROM servers INTERSECT SELECT srvr_id FROM serv_inst; -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 141 | 4560 | 6 (84)| | 1 | INTERSECTION | | | | | -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 141 | 4560 | 6 (84)| | 1 | INTERSECTION | | | | | Database 11gR2 The SQL Statement Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 49 | 1 | INTERSECTION | | | | | | 2 | SORT UNIQUE NOSORT | | 141 | 564 | 2 (50)| | 3 | INDEX FULL SCAN | PK_SERVERS | 141 | 564 | 1 (0)| | 4 | SORT UNIQUE | | 999 | 3996 | 4 (25)| | 5 | INDEX FAST FULL SCAN| IX_SERV_INST | 999 | 3996 | 3 (0)| -------------------------------------------------------------------------- | 1 | INTERSECTION | | | | | | 2 | SORT UNIQUE NOSORT | | 141 | 564 | 2 (50)| | 3 | INDEX FULL SCAN | PK_SERVERS | 141 | 564 | 1 (0)| | 4 | SORT UNIQUE | | 999 | 3996 | 4 (25)| | 5 | INDEX FAST FULL SCAN| IX_SERV_INST | 999 | 3996 | 3 (0)| -------------------------------------------------------------------------- --------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| --------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 141 | 4560 | 20 (10)| | 1 | INTERSECTION | | | | | | 2 | SORT UNIQUE | | 141 | 564 | 10 (10)| | 3 | TABLE ACCESS FULL | SERVERS | 141 | 564 | 9 (0)| | 4 | SORT UNIQUE | | 999 | 3996 | 10 (10)| | 5 | TABLE ACCESS FULL | SERV_INST | 999 | 3996 | 9 (0)| --------------------------------------------------------------------------- --------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| --------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 141 | 4560 | 20 (10)| | 1 | INTERSECTION | | | | | | 2 | SORT UNIQUE | | 141 | 564 | 10 (10)| | 3 | TABLE ACCESS FULL | SERVERS | 141 | 564 | 9 (0)| | 4 | SORT UNIQUE | | 999 | 3996 | 10 (10)| | 5 | TABLE ACCESS FULL | SERV_INST | 999 | 3996 | 9 (0)| --------------------------------------------------------------------------- SQL> SELECT table_name, blocks 2 FROM user_tables 3* WHERE table_name IN ('SERVERS', 'SERV_INST'); TABLE_NAME BLOCKS ------------------------------ ---------- SERVERS 28 SERV_INST 28 SQL> SELECT table_name, blocks 2 FROM user_tables 3* WHERE table_name IN ('SERVERS', 'SERV_INST'); TABLE_NAME BLOCKS ------------------------------ ---------- SERVERS 28 SERV_INST 28 Database 12gR1
- 50. This Adds To The Confusion SQL> SELECT blocks 2 FROM dba_tables 3 WHERE owner = 'UWCLASS' 4 AND table_name = 'SERVERS'; BLOCKS ---------- 28 SQL> SELECT blocks 2 FROM dba_segments 3 WHERE owner = 'UWCLASS' 4 AND segment_name = 'SERVERS'; BLOCKS Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 50 BLOCKS ---------- 32
- 51. This Makes It Even Worse SQL> DECLARE 2 uf NUMBER; 3 ub NUMBER; 4 f1 NUMBER; 5 f1b NUMBER; 6 f2 NUMBER; 7 f2b NUMBER; 8 f3 NUMBER; 9 f3b NUMBER; 10 f4 NUMBER; 11 f4b NUMBER; 12 fbl NUMBER; 13 fby NUMBER; 14 BEGIN 15 dbms_space.space_usage('UWCLASS','SERVERS', 'TABLE', uf, ub, f1, f1b, f2, f2b, f3, f3b, f4, f4b, fbl, fby); 16 17 dbms_output.put_line('unformatted blocks: ' || TO_CHAR(uf)); 18 dbms_output.put_line('unformatted bytes: ' || TO_CHAR(ub)); 19 dbms_output.put_line('blocks 0-25% free: ' || TO_CHAR(f1)); 20 dbms_output.put_line('bytes 0-25% free: ' || TO_CHAR(f1b)); 21 dbms_output.put_line('blocks 25-50% free: ' || TO_CHAR(f2)); 22 dbms_output.put_line('bytes 25-50% free: ' || TO_CHAR(f2b)); Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 51 22 dbms_output.put_line('bytes 25-50% free: ' || TO_CHAR(f2b)); 23 dbms_output.put_line('blocks 50-75% free: ' || TO_CHAR(f3)); 24 dbms_output.put_line('bytes 50-75% free: ' || TO_CHAR(f3b)); 25 dbms_output.put_line('blocks 75-100% free: ' || TO_CHAR(f4)); 26 dbms_output.put_line('bytes 75-100% free: ' || TO_CHAR(f4b)); 27 dbms_output.put_line('full blocks: ' || TO_CHAR(fbl)); 28 dbms_output.put_line('full bytes: ' || TO_CHAR(fby)); 29 END; 30 / unformatted blocks: 16 unformatted bytes: 131072 blocks 0-25% free: 0 bytes 0-25% free: 0 blocks 25-50% free: 1 bytes 25-50% free: 8192 blocks 50-75% free: 0 bytes 50-75% free: 0 blocks 75-100% free: 11 bytes 75-100% free: 90112 full blocks: 0 full bytes: 0 only 12 blocks are formattedonly 12 blocks are formatted
- 52. AWR Difference Reports SQL scripts located at $ORACLE_HOME/rdbms/admin File Name Strategy awrddrpi.sql Report on differences between differences between values recorded in two pairs of snapshots. This script requests the user for the dbid and instance number of the instance to report on, for each snapshot pair, before producing the standard report. awrddrpt.sql Defaults the dbid and instance number to that of the current instance connected-to, then calls awrddrpi.sql to produce the Compare Periods report. awrextr.sql SQL/Plus script to help users extract data from the AWR. awrgdrpi.sql RAC Version of Compare Period Report. awrgdrpt.sql This script defaults the dbid to that of the current instance connected-to, defaults instance list to all available instances and then calls awrgdrpi.sql to produce the Workload Repository RAC Compare Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 52 available instances and then calls awrgdrpi.sql to produce the Workload Repository RAC Compare Periods report. awrgrpt.sql This script defaults the dbid to that of the current instance connected-to, then calls awrgrpti.sql to produce the Workload Repository RAC report. awrgrti.sql SQL*Plus command file to report on RAC-wide differences between values recorded in two snapshots. This script requests the user for the dbid before producing the standard Workload Repository report.
- 53. AWR Difference Reports SQL scripts located at $ORACLE_HOME/rdbms/admin File Name Strategy awrinfo.sql Outputs general AWR information such as the size, data distribution, etc. in AWR and SYSAUX. The intended use of this script is for diagnosing abnormalities in AWR and not for diagnosing issues in the database instance. awrrpt.sql Defaults the dbid and instance number to that of the current instance connected-to, then calls awrrpti.sql to produce the report. awrrpti.sql SQL*Plus command file to report on differences between values recorded in two snapshots. This script requests the user for the dbid and instance number of the instance to report on, before producing the standard report. awrsqrpi.sql SQL*Plus command file to report on differences between values recorded in two snapshots. This script Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 53 awrsqrpi.sql SQL*Plus command file to report on differences between values recorded in two snapshots. This script requests the user for the dbid, instance number and the sql id, before producing a report for a particular sql statement in this instance. awrsqrpt.sql Defaults the dbid and instance number to that of the current instance connected-to then calls awrsqrpi.sql to produce a Workload report for a particular sql statement. awrupd12.sql This script updates AWR data to version 12c. It only modifies AWR data that has been imported using awrload.sql, or data from before changing the database DBID. In other words, it doesn't modify AWR data for the local, active DBID.
- 54. AWR Difference Reports Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 54
- 55. AWR Difference Reports Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 55
- 56. AWR Difference Reports Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 56
- 57. Summary Do not guess If your problem is not caused by slow hardware then faster hardware will just create a problem faster Gather information from a date-time range when everything was good and compare with the date-time range when the problem was observed If the problem is slow SQL ... is the SQL statement itself slow or is it the entire environment that should be examined? Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 57 entire environment that should be examined? If the problem is related to CPU ... do you need more/faster CPUs or a better design? If the problem is related to I/O ... do you need a new SAN or better indexing? If the problem is related to network bandwidth ... do you need new NIC cards and to stop network virtualization? Yes you probably do
- 58. Conclusion Do not tune by guessing Do not blame everything on bad SQL statements Until you can prove that the root cause is the SQL Use the scientific method Construct a hypothesis Test the hypothesis by doing experiments Validate your conclusion Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 58 Validate your conclusion Only rewrite SQL after you have established conclusively that the root cause is the way a statement has been written and that rewriting the SQL statement will address ALL issues You will most likely rewriting very few SQL statements
- 59. ERROR at line 1: ORA-00028: your session has been killed Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 59 Contact: [email protected] +1 206-669-2949 Skype: damorgan11g
- 60. Thank You Daniel A. Morgan | [email protected] | www.morganslibrary.org Not SQL Performance Tuning Presented: APAC Tokyo - June, 2015 60 Thank You Contact: [email protected] +1 206-669-2949 Skype: damorgan11g