migrated 13 tb ebs production to 5 node 11gr2 rac on...
TRANSCRIPT
Migrated 13 TB EBS Production to 5 node 11gR2 RAC on vSphere - A Deep Dive
Charlene Huang, Database Architect Eisen Kung, Senior Manager
2
Introduction
100% Virtualized
Oracle Architecture
Best Practices
Migration and Testing
Reporting and Backup Solution
Virtualization Benefits
Summary and Next Steps
Reference
Q&A
Agenda
Introduction
VMware IT – 100% Virtualized
Why Run Oracle RAC on vSphere?
Lower capital and operating costs
Maximize the use of hardware resources without compromising application
performance
Achieve 100% virtualization in IT Production footprint using VMware products
A major step toward Cloud-ready architecture
Increase agility, efficiency, scalability, and availability to support VMware rapid
growth
Efficient provisioning of new RAC cluster using VM template and Oracle cloning
Simplify Disaster Recovery (DR) solution using VMware Site Recovery
Manager (SRM)
Success Criteria
No compromise in performance after the transition from physical
to virtual
Minimal service disruption during the transition from physical to
virtual
Backup and reporting must be offloaded from production and
continue to meet current SLA
No reduction in resiliency
Must enhance/simplify current disaster recovery solution
Establishing a Baseline
• A snapshot of the physical code base which passed the functional,
performance, and resiliency test rigor was taken
• Use above code base to build the virtual environment
• Took month-end and quarter-end timings from the physical for the
Backup and Reporting Refreshes
VMware IT – Oracle Architecture
Oracle EBS Virtual Solution
Oracle EBS RAC Virtual Solution
Service APP SOA Portal CCM1 CCM2 CCM3 DCO
Preferred Node Node1
Node2 Node3 Node4 Node5
Node3
Node4
Node5
Available Node Node3
Node4
Node4
Node5 Node3
Node5
Node3
Node4
Application EBS
Web/Forms
Fusion
Middleware Portal
Concurrent Manager
Server 1
Concurrent Manager
Server 2
Concurrent Manager
Server 3
Oracle
Discoverer
Multiple custom services are defined in RAC to distribute the load across different nodes.
Node1 and 2 primarily handle OLTP; Node3, 4, and 5 handle batch jobs and reports.
OLTP Batch/Reporting
EBS Infrastructure Details
11
Database is a 5 node RAC cluster
• Oracle Linux Server release 6.2 with 24 vCPU and 256GB memory
• Oracle Grid infrastructure 11.2.0.3.0 64bit; Oracle Database 11.2.0.3.0 64bit
• Virtual RDM is used for database storage
• VMDK is used for OCR and Voting disks
• Jumbo Frames are used for Oracle interconnect
Three node concurrent manager servers
• Oracle Linux Server release 5.7 32 Bit with 8 vCPU and 32GB memory
Six node internal and one node external application server
• Oracle Linux Server release 5.7 32 Bit with 2 vCPU and 8GB memory
Three node Oracle Discoverer servers (10.1.2.55.26)
• Oracle Linux Server release 5.7 64 Bit with 2 vCPU and 4GB memory
EBS Infrastructure Details (cont)
12
Two node RAC Active Data Guard
• Oracle Linux Server release 6.2 64 bit with 12 vCPU and 64GB memory
• Oracle Grid infrastructure 11.2.0.3.0 64 bit; Oracle Database 11.2.0.3.0 64bit
• Virtual RDM is used for database storage
• VMDK is used for OCR and Voting disks
• Jumbo Frames are used for oracle interconnect
Reporting Environment
• Single node Database
o Oracle Linux Server release 6.2 64 bit with 12 vCPU and 64GB memory
• 3 node App and Discoverer servers
o Oracle Linux Server release 5.7 32 Bit with 4 vCPU and 8GB memory
• 1 node concurrent manager
o Oracle Linux Server release 5.7 32 Bit with 4 vCPU and 8GB memory
Backup Environment
• Oracle Linux Server release 6.2 64 Bit with 12 vCPU and 64GB memory
Server and Hardware Diagram
13
Storage Diagram
14
ESXi Cluster
Dual Fabric SAN
PowerPath/VE Multi-Pathing
VMAX 40
With FAST VP
VPLEX RecoverPoint / SRM
VPLEX Quad Engine VS2
UID
HP ProLiantBL660c Gen8
UID
HP ProLiantBL660c Gen8
UID
HP ProLiantBL660c Gen8
UID
HP ProLiantBL660c Gen8
UID
HP ProLiantBL660c Gen8
UID
HP ProLiantBL660c Gen8
Hardware Configuration
15
Server Hardware
• 6 HP Blade BL660c Gen 8 Server
4 Sockets x 8 Cores (32 Cores, 64 Threads), 512 GB DDR3 Memory, 2 x 2-Port FlexibleLOM
• 2 HP Virtual Connect Flex-10/10D Modules
• Dual-Port 8Gb QLogic HBA
Storage Hardware (Not Dedicated)
• SAN Fabric: Brocade DCX 8510 (Dual-Fabric)
• EMC VPLEX VS2 Virtualized Storage (Front End)
• EMC Symmetrix VMAX 40K (Backend)
• FAST VP
VMware IT – Best Practices
Best Practice on vSphere Implemented
17
BIOS setting
• Disable power management
• Enable hyper-threading
• Enable hardware-assisted virtualization features (VT-x, EPT)
Memory setting
• Use huge memory pages in guest OS to reduce TLB overhead
(Configure HugePages the same as SGA size).
• Avoid an expensive swapping overhead in guest OS (Set
vm.swappiness = 0 in /etc/sysctl.conf)
• Memory Reservation - Reserve memory equal to the size of the Oracle
SGA, PGA, plus background processes
Best Practice on vSphere Implemented (cont)
18
Network
• Use VMXNET3 network adapter type
• Enable Jumbo Frame for the RAC Heartbeat Link
• Disable dynamic coalescing to reduce latency on the network links for
all vNICs (example: ethernetX.coalescingScheme = disabled)
IO related setting
• Use paravirtualized SCSI Controller (PVSCSI)
• Use "NOOP" IO scheduler per block device in guest OS
• Increase the limit on the maximum number of outstanding disk requests
per device
Best Practice on vSphere Implemented (cont)
19
Disk related configuration
• Virtual RDM
Disk creation only needs to be done once for the 1st VM in the RAC. The rest of
the VMs will reference the same VMDK pointer
Disk creation can be done either through the vSphere client or the ESXi shell
• VMDK
Shared VMDK disks need to be Thick Provision Eager Zeroed virtual disks
• Enable Multi-Writer for shared disk
• Set SCSI Bus Sharing to None
• Enable Disk UUID
VMware IT – Migration and Testing
Migrating to Virtual RAC with Minimal Downtime
21
• Install 11.2.0.3 Grid and database software on 5 node cluster on VM
• Apply the latest PSU patch
• Allocate the same amount of storage to 5 node RAC virtual cluster
• Use EMC TimeFinder clone to sync the data from physical to virtual
• Setup Data Guard on virtual RAC nodes
• During downtime, perform Data Guard switchover (30min database
downtime)
• Clean up FND nodes and run autoconfig on new database servers
• For CCM and application servers, shared appl_top is used. During
downtime, mount the same NAS share to new VMs, then follow Oracle
support note 233428.1 to complete the migration
Functional Testing (1)
22
Test Coverage Type of Test Portals considered in Execution
EBS Internal Modules EBS (Automation/Manual) testing Execute the existing EBS regression
suite.
SC,OM,AR,PO,AP,GL,CM,FA,VPP,
Eclipse,PER VM,Keystone, Niagara,
Term Lic.
3rd Party Integration
(which will interact w/
EBS System )
EBS Manual execution covering
major scenarios
EDI, Iexpenses, Bookings 2.0,Trade
Beam,Change Point,Blackline,
Discoverer Reports, OM/SC Reports,
EDI Conc. Program Reports & Email
validations, Workflow Emails, Cyber
source – Credit card Order
Functional Testing (2)
23
Test Coverage Type of Test Portals considered in Execution
End to End Smoke
Test for customer
facing portals that
interact with EBS DB
Portal (Automation/Manual)
Testing covering P1 test
scenarios only
EMS,ISV,OEM, DR, Admin, Support,
Flex, Downloads, VPP, Partner Eval,
Public EVAL, Beta.Next, Communities,
SCA, Model N
End to End Smoke
Test Scenarios from
SFDC
SFDC Manual Testing
Will be covering Smoke Test
scenarios only
Admin, EMS, NFR Orders, UCM,
Partner Sync, Update Contact, IVR
3rd Party Integration
(which will interact
EBS System )
Manual testing by BA Citi - Iexpense & CFX, Wellsfargo,
FieldGlass , CDW and Dell Punchout,
RevStream, Commissions, GetPaid
Custom.pll – iSupplier, Vertex
Validations, My Learn (PMO), Work
day, BI, E-Trade, Zoomax, HRMS --
User creations, MnA Interfaces
Performance Testing
24
Test load Detail
1XQE (normal day-to-day load
during quarter end)
• 350 Test Cases, 4,677 Atomic Transactions
• 2,000 Concurrent Users, 567 Orders/hour Placed
2XQE (double the day-to-day load
during quarter end)
• 350 Test Cases, 4,677 Atomic Transactions
• 4,000 Concurrent Users, 1,134 Orders/hour Placed
Product Launch (the load that is
experienced during a product
launch)
• 350 Test Cases, 4,677 Atomic Transactions
• 3,000 Concurrent Users, 1,432 Orders/hour Placed
Endurance Test (1X normal day
load, 72 Hours)
• 350 Test Cases, 4,677 Atomic Transactions
• 2,000 Concurrent Users, 567 Orders/hour Placed
Conducted with HP LoadRunner 9.1. Three load tests were executed at different load levels and
one endurance test was executed for 72 hours.
Resiliency Testing
25
• Public NIC Failure/ Interconnect NIC Failure
• Storage path failure (disable one path)
• Network driver failure on master or slave node
• Use vMotion to live migrate all RAC nodes to different hosts
• CRSD/EVMD/CRSDD Process Failure
• RDBMS Process Failure
• ASM Instance failure ( Unplanned)
• Scan Listener/ Listener Failure ( Unplanned )
• Multiple Instance Failure (Unplanned)
• Planned Instance Termination
• Rebooting RAC node/ Power-off one node
• CCM server failure
All test cases were performed under various system loads, with SoapUI PRO calls, HP LoadRunner 9.1 injecting 350 test cases and 4,677 atomic transactions using 2,000 concurrent users) & transactions generated using QTP (Quick Test Professional) Tool, and Mercury Interactive Functional Testing Tool
Utilization Comparison Considerations
26
Physical RAC 5 Node RAC on HP G7 blades, each Node has:
CPU = 20 core with hyper threading
Memory = 256GB
SGA Size = 38GB
Virtual RAC combined with Hardware Tech Refresh 5 Node RAC on HP Gen 8 blades, each Node has:
vCPU = 24
Memory = 256GB
SGA Size = 62GB
RAC Service Configuration OLTP transactions are configured to connect to Node 1, 2
Concurrent programs are configured to connect to Node 3, 4, 5
Discoverer and reporting are configured to connect to Node 3, 4, 5
Memory Utilization Percentage Comparison
27
0
5
10
15
20
25
30
35
40
VMWPROD1 VMWPROD2 VMWPROD3 VMWPROD4 VMWPROD5
Memory Utilization Percentage
Physical ME
Virtual ME
Virtual QE
Physical SGA is 38GB
Virtual SGA is 62GB
DB Single Block Disk Read (ms) Comparison
28
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
VMWPROD1 VMWPROD2 VMWPROD3 VMWPROD4 VMWPROD5
DB Single Block Disk Read (ms)
Physical ME
Virtual ME
Virtual QE
Batch Jobs Performance Comparison
29
0
100
200
300
400
500
600
ContractsStatus Change
by Range
VM AutoRenewal
NotificationTable Script
VM BookingMN.Next
Allocations
VMCommissions
LoadTransactions
VM ModifyUpgradedCovered
Product Lines
VM: KeystoneData Profile
ExtractProgram
VM: SCExceptions
XXRS:revStream
PentupRevenueReport
XXVM RDM ARTransactions
Average Run Time in Minutes
Physical ME
Virtual ME
Virtual QE
Challenges and Resolutions
30
• Network Latency between VMs – affects OEL 6.2 due to LRO setting.
Override the inbox vmxnet3 driver with the VMware driver from VMware
tools
• Maximum allowed outstanding disk requests per device affects IO
throughput. Increase from default 32 to 128
• vMotion with RDM – availability. Able to vMotion using virtual mode
RDM
• CRS Bug – availability – affect OEL 6.x. The maximum process value
of the OHASD bin is too small if the CRS is started manually (Add
“ulimit -u 16384” to ohasd and crswrap.sh.sbs). Refer to Oracle support
Documentation ID 1594606.1 for the detailed resolution
• High storage bandwidth utilization required additional ISLs in the trunk.
Operation Support Awareness
31
• Do not attempt to take a snapshot of any RAC nodes with shared
RDMs; doing so will cause VMs to power off.
• Storage vMotion does not work if there is shared disk.
(http://kb.vmware.com/kb/1005241)
• Storage vMotion works with RDMs when no disks are shared; however,
any custom disk label and folder will be removed
• Hot Add or Delete Shared Disks without downtime (with vSphere API)
VMware IT – Reporting and Backup Solution
Oracle EBS Virtual Solution
2 Node Cluster
App
Discoverer
Async
Redo
Transport
Ad Hoc Query
Active
Data Guard
App3 App5 App6App4App1 App2
CCM2 CCM3CCM1
5 Node Cluster
Storage syncStorage sync
Backup serverReporting server
CCM
App1-3
CCM
SOA
Portal
Reporting Refresh Solution
34
Business Requirement: 2hr downtime window to perform daily refresh of reporting instance end to end.
A master script automatically performs: • Shutdown reporting instance.
• Perform storage sync.
• Recover and open database.
• Rename database to reporting instance name.
• Run autoconfig on database, concurrent manager, and application servers.
• Run custom update scripts to clean up CCM queue, e-mail alerts, etc.
• Bring up concurrent manager services and application services.
Oracle EBS Virtual Solution
2 Node Cluster
App
Discoverer
Async
Redo
Transport
Ad Hoc Query
Active
Data Guard
App3 App5 App6App4App1 App2
CCM2 CCM3CCM1
5 Node Cluster
Storage syncStorage sync
Backup serverReporting server
CCM
App1-3
CCM
SOA
Portal
Backup Solution
36
Full backup on the dedicated backup server
• Install single-node 11.2.0.3 Grid infrastructure and database software.
• Allocate same amount of storage as production server.
• A master script automatically performs
storage sync from production host to backup host,
Mount Database
Perform the RMAN backup to Data Domain using EMC Networker.
13TB database completed in 5 hours
Archivelog backup twice daily on primary RAC nodes using EMC Networker.
Disaster Recovery Solution using VMware Site Recovery Manager (SRM)
37
VMware vSphere
VMware
vCenter Server
Site Recovery
Manager
VMware
vCenter Server
Site Recovery
Manager
VMware vSphere
Site A – Santa Clara, CA (Primary) Site B – Wenatchee, WA (Recovery)
Servers Servers
EMC RecoverPoint
• Recovery site is located 800 miles
Away from Primary site
• EMC RecoverPoint Replication
leveraging EMC VPLEX
RecoverPoint splitter
• Cisco OTV deployed to extend
Layer 2 network to the Recovery
datacenter
• Simplified recovery using VMware
vCenter Site Recovery Manager
VMware IT – Virtualization Benefits
Reduced Datacenter Footprint by 60%
39
# of
Blades
# of
chassis EBS RAC Reporting Active
Data
Guard
CCM SOA/IDM Backup
Physical 15 2 5 node
20 core
256G mem
1 node
12 core
96G mem
3 nodes
8 core
64G mem
2 nodes SOA
2 nodes IDM
12 core
96G mem
1 node
12 core
96G mem
Virtual 6 1 5 node
24 vcpu
256G mem
1 node
12 vcpu
64G mem
2 nodes
12 vcpu
64G mem
3 nodes
8 vcpu
32G mem
3 nodes SOA
3 nodes IDM
8 vcpu
32G mem
1 node
12 vcpu
64G mem
Reduced Physical Servers and Oracle Licensing
Products / Features Used
40
• VMware vMotion – VM mobility enables us to do online maintenance and
infrastructure upgrade.
• VMware DRS – provides automated resource balancing based on real time
resource usage. Anti-affinity rules deployed to achieve system distribution over
multiple hosts in an application cluster.
• VMware HA – automated system restart/recovery when host failure/isolation
occurs in the cluster.
• Hot-plug for CPU, memory and storage.
• SRM – used along with EMC RecoverPoint technology for disaster recovery
solution.
• vSphere API for automated disk management.
• vCenter Operations Manager for vSphere monitoring
VMware IT – Summary
Summary
42
Success Criteria AccomplishmentCombined the RAC migration to virtual with the
Hardware Tech Refresh
No compromise in performance after the transition
from physical to virtual.
Virtual EBS 5 node RAC database performance
equaled or better its physical predecessor
Minimal service disruption during the transition
from physical to virtual.
The downtime of the switchover from physical to
virtual was reduced to 30 mins.
Backups and reporting must be offloaded from
production and continues to meet current SLA
With the tech refresh, the 13TB database backup
was reduced from 11 hrs to 5 hrs
Reporting refresh was within the 2 hour SLA
window
No reduction in resiliency. Increased resiliency with addition of VMHA
Must have a disaster recovery solution. Achieved a unify disaster recovery solutio
Reference
43
• VMware vSphere distributed Switch Best Practices http://www.vmware.com/resources/techresources/10250
• VMFS Heap, PB Cache Slab, and Max Open Addressable Space: vSphere 2013 and Beyond https://wiki.eng.vmware.com/VMFS/AddrModesPBCacheAndHeapSizes#VMFS_Heap.2C_PB_Cache_Slab.2C_and
_Max_Open_Addressable_Space:_vSphere_2013_and_Beyond
• Disabling simultaneous write protection provided by VMFS using the multi-writer flag
(1034165) http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1034165
• Oracle DB on VMware Best Practices http://www.vmware.com/files/pdf/partners/oracle/Oracle_Databases_on_VMware_-_Best_Practices_Guide.pdf
• Oracle DB on VMware RAC deployment Guide http://www.vmware.com/files/pdf/solutions/oracle/Oracle_Databases_VMware_RAC_Deployment_Guide.pdf
• vSphere 5.5 performance best practices http://www.vmware.com/pdf/Perf_Best_Practices_vSphere5.5.pdf
• Tuning VM for latency-sensitive workload http://www.vmware.com/files/pdf/techpaper/VMW-Tuning-Latency-Sensitive-Workloads.pdf
Reference (cont)
44
• Pre-installation requirements for 11gR2 http://download.oracle.com/docs/cd/E11882_01/install.112/e10812/prelinux.htm
• Installing 11g R2 Grid infrastructure http://docs.oracle.com/cd/E11882_01/install.112/e22489/crsunix.htm#CWLIN315
• Oracle CRS Bug 17301761 THE MAXIMUM PROCESS VALUE OF OHASD,BIN IS TOO
SMALL IF THE CRS STARTS MANUALLY Oracle support Documentation ID 1594606.1
• Adding a EBS application node when using a shared file system oracle support note 233428.1 section 5
Acknowledgement
Zaigui Wang -- IT vSphere Architect Kandy O'Mara -- IT Storage Architect Chanh Chi -- Director – IT Cloud Operations Sudhir Balasubramanian -- CoE Solutions Architect
Q&A