maximizing performance on oracle 11g se rac
DESCRIPTION
Maximizing performance on Oracle 11g SE RAC. Wendy Chen, Systems Engineer Naveen Iyengar, Systems Engineer Global Solutions Engineering, Dell Inc. DELL CONFIDENTIAL. SE vs. EE. Oracle 11g database is available in multiple editions Enterprise Edition (EE) Single or clustered servers - PowerPoint PPT PresentationTRANSCRIPT
DELL CONFIDENTIAL
MAXIMIZING PERFORMANCE ON ORACLE 11G SE RAC
Wendy Chen, Systems EngineerNaveen Iyengar, Systems EngineerGlobal Solutions Engineering, Dell Inc.
SE VS. EE
Oracle 11g database is available in multiple editions Enterprise Edition (EE)
– Single or clustered servers– No limit on the maximum number of CPU (up to the maximum
number of nodes supported in a RAC cluster)– Contains all Oracle database components, and can be further
enhanced with the purchase of options and packs Standard Edition (SE)
– Single or clustered servers– Up to a maximum of 4 CPU sockets in the single or clustered
servers– Includes selected, but not all the features come with EE – Built from the same code base as EE– Ideally suited to the needs to SMB with enterprise level
performance
FEATURE AVAILABILITY
A number of features that come with Oracle 11g EE are not available in SE. For example,– Data guard– Rolling upgrades– Online index and table organization– Parallel backup and recovery– Tablespaces point-in-time recovery– Flashback table / transaction / database– Parallel query / statistics gathering / index builds / data pump
export and import– Transportable tablespaces– Infiniband support
Before deploying SE, make sure that you do not require any of the non-supported features.
PAY-AS-YOU-GROW SCALABILITY
“Pay-as-you-grow” methodology to scale up to four single socket machines
Storage Group
RAID 10
Flash Recovery Area Volume
Storage Group
OCR and Voting Disk Volume
Data VolumeStorage Pool
RAID 10 RAID 10
Client Systems
Dell Gigabit Ethernet Switches for Oracle Cluster Private Network
Dell PowerEdge R610 Servers
Dell PowerConnect 6248 Gigabit Ethernet Switches* for
iSCSI Storage Area Network
Dell EqualLogic PS4000XV iSCSI Storage Arrays
TESTING PERFORMANCE CAPABILITIES OF 11G SE RAC
Quest Benchmark Factory TPC-C– A single node Oracle 11.1.0.7 SE database: 100 to 10,000 users
– Two-node Oracle 11.1.0.7 SE RAC database: 100 to 10,000 users
– Three-node Oracle 11.1.0.7 SE RAC database: 100 to 10,000 users
– Four-node Oracle 11.1.0.7 SE RAC database: 100 to 10,000 users
Server Up to four Dell PowerEdge R610 servers with:A single Intel® Xeon® X5560 quad-core 2.80 GHz CPU with 8M cache, 6.40 GT/s QPI and TURBO and HT mode enabled
24 GB of RAM or 48 GB of RAM, with 4GB or 8GB DIMMs respectivelyFour 1Gb Broadcom NetXtreme II NIC ports for iSCSI traffic
External Storage Two DellTM EqualLogicTM PS4000XV iSCSI storage arrays, each with 16 15K RPM 146GB SAS hard drives
Volume Configuration Three 220 GB volumes for database files; One 150 GB volume for Flash Recovery area; One 2 GB volume for OCR, CSS, and SPFILE
OS and Device Driver Oracle Enterprise Linux 5 Update 3Open iSCSI initiator iscsi-initiator-utils-6.2.0.868-0.18.el5Device Mapper multipath driver device-mapper-multipath-0.4.7-23.el5
Storage Network Two stacked Dell PowerConnect 6248 gigabit Ethernet switches for iSCSI SAN
Test Software Quest Benchmark Factory 5.7.1 Oracle 64 bit 11.1.0.7 SE RAC
Database Configuration Up to four Oracle 11.1.0.7 SE RAC with:13 GB memory_target on each instance
PERFORMANCE MONITORING
CPU utilization: SAR (System Activity Report)
nohup sar 1 30000 > ~/sar-4nodes-run195-node1.txt & CPU %user %nice %system %iowait %steal %idle
02:31:43 PM all 0.38 0.00 0.25 4.25 0.00 95.1202:31:44 PM all 0.25 0.00 0.25 4.12 0.00 95.3802:31:45 PM all 0.12 0.00 0.50 0.25 0.00 99.1302:31:46 PM all 0.62 0.00 0.62 1.12 0.00 97.6202:31:47 PM all 0.87 0.00 1.12 1.99 0.00 96.0102:31:48 PM all 0.13 0.00 0.38 5.63 0.00 93.8702:31:49 PM all 0.62 0.00 0.37 1.87 0.00 97.13...02:33:29 PM all 0.38 0.00 0.12 0.25 0.00 99.2502:33:30 PM all 0.00 0.00 0.25 0.00 0.00 99.7502:33:31 PM all 0.25 0.00 0.62 0.25 0.00 98.8802:33:32 PM all 0.37 0.00 0.25 0.00 0.00 99.3802:33:33 PM all 0.50 0.00 0.50 1.13 0.00 97.8702:33:34 PM all 0.25 0.00 0.25 0.25 0.00 99.25
PERFORMANCE MONITORING
Physical memory utilization: OS Watcher
nohup ./startOSW.sh 5 96 &
zzz ***Sun Jun 21 23:04:59 CDT 2009MemTotal: 49434036 kBMemFree: 34270564 kBBuffers: 354892 kBCached: 12637136 kBSwapCached: 7732 kBActive: 10278772 kBInactive: 4483804 kBHighTotal: 0 kBHighFree: 0 kBLowTotal: 49434036 kBLowFree: 34270564 kBSwapTotal: 8388600 kBSwapFree: 8239712 kB
PERFORMANCE MONITORING
Storage utilization: EqualLogic SAN HeadQuarters
TUNING ORACLE MEMORY SIZE The optimal size of SGA is the point where the marginal benefit of physical
I/O reduction begins to decline The optimal size of PGA is when the estimated PGA over-allocation size is 0
SGA Target Size (M) SGA Size Factor Est DB Time (s) Est Physical Reads
3,168 0.38 1,327,870 37,779,644
4,224 0.50 895,776 20,459,187
5,280 0.63 728,220 13,742,069
6,336 0.75 641,710 10,272,675
7,392 0.88 598,910 8,558,231
8,448 1.00 569,142 7,364,453
9,504 1.13 549,509 6,576,457
10,560 1.25 534,597 5,979,936
11,616 1.38 524,011 5,555,743
12,672 1.50 519,685 5,381,942
13,728 1.63 516,046 5,236,863
14,784 1.75 516,112 5,236,863
15,840 1.88 516,117 5,236,863
16,896 2.00 516,119 5,236,863
PGA Target Est (MB)
Size FactrW/A MB
Processed
Estd Extra W/A MB Read/
Written to Disk
Estd PGA Cache Hit %
Estd PGA Overalloc Count
Estd Time
912 0.25 564.98 15.25 97.00 113 1,740
1,824 0.50 564.98 15.25 97.00 77 1,740
2,736 0.75 564.98 15.25 97.00 50 1,740
3,648 1.00 564.98 0.00 100.00 17 1,694
4,377 1.20 564.98 0.00 100.00 0 1,694
5,107 1.40 564.98 0.00 100.00 0 1,694
5,836 1.60 564.98 0.00 100.00 0 1,694
6,566 1.80 564.98 0.00 100.00 0 1,694
7,296 2.00 564.98 0.00 100.00 0 1,694
TESTING SE RAC SCALABILITY WITH 24 GB RAM PER NODE – CPU UTILIZATION
TESTING SE RAC SCALABILITY WITH 24 GB RAM PER NODE – MEMORY AND STORAGE UTILIZATION
TESTING SE RAC SCALABILITY WITH 24 GB RAM PER NODE – OVERALL PERFORMANCE ANALYSIS
As the systems near their maximum memory performance, they progressively utilize more swap space and the database starts to get unstable.
When comparing the CPU graph with the memory graph, at the maximum user load there is still ample CPU capability even once the test fails.
Given that the Oracle memory area (memory_target) is set to 13 GB to minimize physical I/O, even at the maximum 4-node cluster size the storage members were delivering IOPS well below their potential, suggesting that the storage disks were not the bottleneck for the overall performance.
TESTING SE RAC SCALABILITY WITH 48 GB RAM PER NODE – CPU UTILIZATION
TESTING SE RAC SCALABILITY WITH 48 GB RAM PER NODE – MEMORY AND STORAGE UTILIZATION
TESTING SE RAC SCALABILITY WITH 48 GB RAM PER NODE – OVERALL PERFORMANCE ANALYSIS
As the physical memory increased from 24 GB to 48 GB, the system performance was no longer limited
by memory, thus the database was able to handle more transactions at higher user loads. Consequently,
the PS4000XV members processed more I/O requests and showed much improved IOPS results
comparing to the 24 GB configuration.
The corresponding disk latency performance was still within the acceptable range of 20 ms in all test
durations of the 48 GB memory configurations, except towards the end of run on the 4-node RAC, when
disk latency started to have a delayed response of above 20 ms. This indicates that the PS4000XV
showed signs of stress after approximately 7000 user loads on the 4-node RAC.
PERFORMANCE CHARACTERISTICS SUMMARY
The 48 GB Oracle RAC cluster scales better than its 24 GB counterpart. For the 24 GB configuration, the TPS
measure scales in the range of 40% - 60% with the addition of each node. With 48 GB per node, the results
demonstrate a near linear scalability, in the range of 75%-95% and that the benefit of additional memory became
more apparent with each additional node.
With 48 GB of physical memory per node, both the CPU and memory resources become limited almost
simultaneously. In an ideal situation, all the resources would be constrained simultaneously. Extra CPU
capability without complementary memory may not lead to ideal scaling characteristics, and likewise extra
memory without CPU capability may be unnecessary.
By adequately balancing memory, CPU and storage you can help make the most out of scaling out an Oracle
RAC cluster. A bottleneck in any one component such as storage, CPU, or memory may result in non-optimal
scaling. Likewise, scaling up of one component when another component is the bottleneck for the system may
not add performance and may be unnecessary.
RESOURCES
Dell whitepaper: Maximizing Performance on a Cost Effective Oracle 11g SE RAC Database on Dell EqualLogic PS4000XV iSCSI Storage http://www.dell.com/downloads/global/solutions/Oracle_SE_RAC.pdf?c=us&cs=555&l=en&s=biz
Dell Oracle Solutions Engineering http://www.dell.com/oracle
Q & A