Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. |
Database Consolidation with Oracle Exadata Techniques and Best Practices
Manish Shah Sr. Principal Product Manager, Exadata Oracle Corporation Ajit Singh Director, Data Technology Platforms New York Stock Exchange (NYSE: ICE)
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Program Agenda
Why Consolidate On Exadata
How To Be Successful With Consolidation On Exadata
Customer Success Story - NYSE
1
2
2
3
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Program Agenda
Why Consolidate On Exadata
How To Be Successful With Consolidation On Exadata
Customer Success Story - NYSE
1
2
3
3
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 4
Odds Of dying in a lifetime*
1 in 98 1 in 7178
*2008 National Safety Council Report
4 People 250 People
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Redundant Database Servers
Redundant Storage
Redundant Connectivity
Redundant and hot swappable power supply
Redundant Components Continuous Application Availability through hardware
Faults
5
• FANs • PDUs • Batteries • Switch • Disk • Flash • Motherboards • DIMMs
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Reduced HA Brownouts • Fast node detection and failover • Automatic detection of Exadata storage failures with low application impact • Automatic detection of Exadata network failures with low application impact
Data Protection • Automatic Hard Disk scrub and repair • Exadata H.A.R.D
Quality Of Service • I/O Latency capping for read and write operations • Exadata Cell I/O Timeout Threshold • Health Factor for predictive failed disk • Identification of underperforming disks and automatic removal • I/O Resource Management • Network Resource Management
Overview of Exadata Specific HA Features
6
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Performance
• Exadata Smart Flash Logging • Active Bonding Network • Exadata Smart Write Back Flash Cache Persistent After Cell Restarts
Management
• Flash and Disk Life Cycle Management Alerts • Cell Alert Summary • LED Notification for Storage Server Disk Removal • Drop Hard Disk for replacement • Drop BBU for replacement • Minimize or eliminate False disk failures
7
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
0
50
100
150
200
40 DB 160 DB
x86
X86
Exadata
How many more databases can be consolidated on
Exadata?
177 ms 132 ms
4X Density With Faster Response
Time
8
Avg. Response Time
0
50
100
150
200
250
300
168 non-CDB
252 PDB
PDB
50% Higher Density
How many more PDB can be consolidated than non-
CDB?
212 ms 14 ms
Avg. Response Time
Exadata Non-CDB
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
The Exadata “Community Effect”
• Standard Technologies • Standard Configurations • Standard Integration • Standard Tuning • Standard Support
9
Exadata Takes Standardization to the Next Level
Server Vendor A
Storage Vendor B
Network Vendor C
Database Vendor D
OS Vendor E
VM Vendor F
The New Global Standard A Company Standard vs
Oracle Public Cloud Oracle Development & Support
1,000’s of Customers and Partners
The Scalability of Telecom, The Availability of Banking, The Security of Government
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Program Agenda
Why Consolidate On Exadata
How To Be Successful With Consolidation On Exadata
Customer Success Story - NYSE
1
2
10
3
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 11
8 Key Success Factors
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 12
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Availability Service Levels for Unplanned and Planned Outages Oracle MAA Availability Tiers
BRONZE
SILVER
GOLD • Comprehensive HA and Disaster Protection • Recovery in seconds with zero or near-zero data loss
• High Availability (HA) for Recoverable Local Outages • Backups plus redo for Oracle data protection
• Basic Service Restart • Backups plus redo for Oracle data protection
PLATINUM • Zero Outage for Platinum Ready Applications • Zero data loss
13
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Zero Application Outage for Platinum Ready Applications Platinum Tier
Site A Site B
Backups
Backups
Outages masked from applications, in-flight transactions preserved
– Application Continuity Zero data loss failover, LAN or WAN
– Active Data Guard / Far Sync Bi-directional replication and zero
downtime maintenance – Oracle GoldenGate
Online patching for applications – Edition-based Redefinition
Automated workload management – Oracle Global Data Services
Tape
Data Guard sync
Tape
Application Continuity
GoldenGate
RAC RAC
14
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 15
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 16
• What criteria should you use to determine # of CDBs? – Prod v/s Non-prod? – Business Functions? – Business SLAs / Common maintenance window? – All of the above?
• It “Depends” ! –
Determine Strategy to move to Multitenant
How many CDBs should you have?
No. I am not going to say that.
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 17
Goal
25 Prod 75 Non Prod
ABC Company 100 Databases
5 DW 20 OLTP 10 DW 65 OLTP
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 18
25 Prod 75 Non Prod
ABC Company 100 Databases
5 DW 20 OLTP 10 DW 65 OLTP
Metal Tier Prod Non Prod
DW OLTP DW OLTP
Platinum 1 5 Gold 2 10 Silver 2 5 2 20
Bronze 8 45
Total 5 20 10 65
Apply Reference Architecture
Start With Database Categorization
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 19
• Separate CDBs for OLTP and DW workloads
• Separate CDBs for different 12c versions. Limit the number of Oracle database homes to a maximum of 5.
• Separate CDBs for Prod and Non-Prod HA Tier. • Consider separate CDB for PDBs that don’t need flash performance so you can turn
IORM Flash cache and Flash log off for that CDB when needed.
• Keep no more than two variants per patch set (e.g. 12.1.0.2.7 – Apr PSU and 12.1.0.2.10 – July PSU)
Guidelines for Separating PDBs into CDBs
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 20
25 Prod 75 Non Prod
ABC Company 100 Databases
5 DW 20 OLTP 10 DW 65 OLTP
Metal Tier Prod Non Prod
DW OLTP DW OLTP
Platinum 1 5 Gold 2 10 Silver 2 5 2 20
Bronze 8 45
Total 5 20 10 65
Apply CDB Guidelines
Start With Database Categorization Apply Reference Architecture
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 21
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Engineered Systems – Capacity Utilization and Planning (ES-CUP)
Program Details
• Free program offered by Oracle for Exadata Customers
• New Customers
• Analyze existing workload
• recommend Exadata machine configuration
• Existing Customers • Recommend solutions to address current challenges • Provide roadmap for ongoing capacity planning exercise
• Engagement Process
• Contact your regional Oracle Sales Team
22
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Engineered Systems – Capacity Utilization and Planning
Conf-call to review objectives
Document engagement scope
Planning and timeline
Customer to describe near-term growth estimates
ACTI
VITI
ES
Kick-Off Data Collection Analyze Readout
Oracle send data collection scripts
Customer execute data collection scripts and send logs to Oracle
Customer execute Exachk and send logs to Oracle
Oracle to analyze information
Crosscheck with Best Practices
Generate Exadata Capacity Utilization and Planning document
Generate roadmap
Remote or onsite delivery of Exadata Capacity Utilization and Planning
Review objectives Knowledge transfer Documentation
shared
ES-CUP
Step 1 Step 2 Step 3 Step 4
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 24
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Consolidation Workbench
Enterprise Manager
25
25 Prod 75 Non Prod
ABC Company 100 Databases
5 DW 20 OLTP 10 DW 65 OLTP
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 26
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Resources To Be Managed
CPU
Memory
Disk
Flash
Exadata Resource Management
CDB1
CDB2
Instance 2
Instance 2
CDB1
CDB2
Instance 1
Instance 1
P1 P2 P3
P4 P5 P6
Exadata Quarter Rack X5-2 27
P1 P2 P3
P4 P5 P6
Compute Node 1 Compute Node 2
Storage Node 1
Storage Node 2
Storage Node 3
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
CDB1 CDB2
P1 P2 P3 P4 P5 P6
OLT
P
BATC
H
OLT
P
BATC
H
OLT
P
BATC
H
OLT
P
BATC
H
OLT
P
BATC
H
OLT
P
BATC
H
HOW DO I ALLOCATE BETWEEN
HOW DO I ALLOCATE SPACE AND BANDWIDTH BETWEEN
CPU
FLASH
DISK HOW DO I ALLOCATE IO REQUESTS BETWEEN
Exadata Resource Management
28
Storage Node 1
Storage Node 2
Storage Node 3
CDB1
CDB2
Instance 1
Instance 1
P1 P2 P3
P4 P5 P6
Compute Node
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 29
Most Important For Consolidation Purpose
Configured on database Configured on database
Configured on database
Configured on database
Configured on database
Configured on database
Configured on database
Between CDBs
PDBs Within CDBs
Workloads Within PDBs
CPU
FLASH
DISK
Instance Caging CDB Resource
Plan
CDB Resource Plan
CDB Resource Plan
Database Resource Plan
Database Resource Plan
Database Resource Plan
Inter-Database IORM
Inter-Database IORM
Configured on Storage Cell
Configured on Storage Cell
Storage Node 1
Storage Node 2
Storage Node 3
CDB1
CDB2
Instance 1
Instance 1
P1 P2 P3
P4 P5 P6
Compute Node
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 30
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Best practices for Exadata planned maintenance
• Leverage Exachk for simple software planning • Leverage zero downtime software maintenance • Usage Out of Place and Standby First Updating • Leverage Lights Out Patching with notification function • Take advantage of Exadata Engineered defaults
31
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 32
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
OVM On Exadata Use Cases and Benefits
Caveats:
• Exadata is optimized to be a Database Machine. The primary goal of OVM on Exadata is to provide a higher level of Isolation between databases.
• Exadata is not a middleware consolidation platform. Consider PCA* or SuperCluster for that purpose.
Recommendation by Oracle:
• Focus on consolidating database with perhaps some light consolidation of other applications. An example of light consolidation would be something like: "I want to consolidate 30 databases, but I also need a few cores to run some ETL, management, and other tools and don't want to deploy another server just for these" .
OVM on Exadata was created for databases but it is a full fledge feature that can be used for any valid reason a customer wants. It is available at no additional cost.
33
*Private Cloud Appliance
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 34
VM VM
Isol
atio
n
Efficiency
Database 12c Multitenant
Many DBs in one Server
Virtual Machines
Dedicated DB Servers
• VMs have good Isolation but poor efficiency and high management – Separate OS, memory, CPUs, and patching – Isolation without need to trust DBA,
System Admin
• Database consolidation in a single OS is highly efficient but less isolated – DB Resource manager isolation adds no
overhead; must trust admins to configure
• Best strategy is to combine VMs with database native consolidation – Multiple trusted DBs/PDBs in a VM – Few VMs per server to limit overhead
DATAB
ASE IN
-MEM
OR
Y
FINANCE VM
SUPPLY CHAIN
VM
SALES VM
OVM On Exadata Use Cases
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Use Cases:
• Hosting, Cloud, Cross department consolidation
• OVMs are trusted partitions – use for license cost reduction/compliance (e.g. create a VM cluster with few cores to match what you are licensed for – best for Oracle options e.g. partitioning)
• Government compliance for workload isolation – PCI/DSS requirements can be met by using OVM, VLAN Tagging and IB Partitioning.
• Host a small app or a tool that needs isolation
No Additional Cost
SALES
SUPPLY CHAIN
FINANCE
DB 11.2 and 12c X2-2, X3-2, X4-2, X5-2
35
OVM On Exadata Use Cases
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
• Exadata OVM allows licensing of Virtual CPUs (Trusted Partitions)* – At least 40% of usable cores on the machine must be licensed for Oracle software – Cannot use both OVM and CoD to license less than 40% of cores – A server must be virtualized or non-virtualized – not mixed – OVM has no License cost; support is included with premier hardware support
36
* http://www.oracle.com/us/corporate/pricing/partitioning-070609.pdf
Virtualized Virtualized
VM1A 12 cores
(24 vCPU)
VM2A 24 cores
(48 vCPU)
DB EE
Non-Virtualized
DB EE RAC
Adv Security
14 cores
DB EE DB IM
VM1AB 12 cores
(24 vCPU)
VM2B 24 cores
(48 vCPU)
DB EE DB EE RAC
Adv Security
60% CoD
Cluster
RAC RAC
Cluster
OVM On Exadata Configuration Guidelines
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 37
Hardware X2-2 X3-2 X4-2 X5-2 Physical Memory /Node (default/max)
72 GB 144 GB (max)
256 GB 512 GB (max)
256 GB 512 GB (max)
256 GB 768 GB (max)
Min Mem per VM 8 GB min + DB or app memory Max Mem per VM 512 GB Mem Defaults (GB) Small - 16; Medium - 32; Large – 64 (Templates adjustable at install time)
Physical CPU threads/node* 24 32 48 72 Min vCPUs per VM 2 Max vCPUs per VM Physical Threads - 2 Default Template vCPUs Small - 2; Medium - 4; Large – 8 (Templates adjustable at install time)
Total Available Disk Space for all DomU
700 GB 700 GB 1.6 TB 1.6 TB
Minimum Used Disk Space per VM
100 GB Sparse allocated per VM for initial deployment. 90 GB required to run one Oracle database version (8 GB required per DB Version). 10 GB available for any other usage.
OVM On Exadata Configuration Guidelines
*1 thread = 1 vCPU
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. DB
FS
RECO
DATA DBFS
RECO
DATA
OVM Cluster 1
OVM Cluster 2
Node 3 (Virtual) Node 4 (Virtual)
DBFS
RECO
DATA
Node 1 (Physical) Node 2 (Physical)
Physical Cluster
Exadata Half Rack Split into Physical and Virtual Clusters
38
OVM On Exadata Example
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
DBFS
RECO
DATA
DBFS
RECO
DATA
Node 3 (Virtual) Node 4 (Virtual)
Dom 0 Dom 0
OVM Cluster 1
OVM Cluster 2
Dom U1
Dom U2
Dom U1
Dom U2
39
OVM On Exadata Example
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
Database Server: Bare Metal / Physical versus OVM
OVM Database Server
dom0
Exadata (Linux, Xen, fw)
domU-3
Exadata (Linux)
Oracle GI/DB homes domU-2
Exadata (Linux)
Oracle GI/DB homes domU-1
Exadata (Linux)
Oracle GI/DB homes
Bare Metal / Physical Database Server
Exadata (Linux, fw)
Oracle GI/DB homes
No change to Storage Grid, Networking between OVM and Bare Metal
OVM On Exadata Software Architecture Comparison
Copyright © 2015, Oracle and/or its affiliates. All rights reserved.
OVM Patching on Exadata • Must maintain patching on both Dom0 and DomU • Grid Infrastructure and RDBMS patching is handled in the same manner as non-virtualized system
• Exadata Software updates on DomU may include updates to both, Exadata Platform Software and OS or it may include update to any one component
• Dom0 only needs Exadata Software Updates
41
OVM On Exadata Maintenance and Patching
New York Stock Exchange Oracle Exadata – Our Journey
Ajit Singh, Director, Data Technology Platforms
New York Stock Exchange (NYSE) and Intercontinental Exchange (ICE)
NYSE - world’s largest stock exchange by market capitalization ($28 trillion)
NYSE is owned by Intercontinental Exchange (ICE) - Architects of the world’s markets
11 exchanges, 7 clearing houses 12,000 + listed contracts and securities
Diverse markets span futures & options on interest rates, commodities, indexes and FX, as well as equities and equity options.
We transact on average in <200 mics.
DATA ENVIRONMENT
November 17, 2015 44
Data volume: ~50-60 billion transactions per day (at peak)
Data size: ~ 15 TB of data per day (at peak)
All data for equities, ETFs, options markets and the feed to the street
Data for trading system latency, other markets
Data services responsible for managing all the data and data platforms including Exadata
Oracle Database environment on Exadata
November 17, 2015 45
Oracle - primary transactional database platform supporting all NYSE businesses
OLTP and Data Mart environments - 250+ databases, over 500 instances
Total database size - 350+ TB
Daily transaction volume - ~1000 million transactions
Large transactional databases: ~10-20 TB databases generating about ~ 300 GB of log per day
Large Data Mart: >100 TB in size (after HCC compression for historical data)
CONSOLIDATION TO EXADATA
2004: 2006: 2014:
17 Exadata – 12 for NYSE
NYSE Exadata footprint
EXADATA FOR CONSOLIDATION
November 17, 2015 49
Some examples: Dev: ~60 databases consolidated on a full rack of X3-2 with an average size of 1.6 TB QA: ~70 QA consolidated on to a full rack of X3-2
with an average size of 2 TB
Prod: ~ 50 databases consolidated on a full rack of X3-2 with an average size of 9.5 TB Mix of OLTP and data marts
Leveraging exadata for performance
November 17, 2015 50
Performance gains of 2x – 25x
Trade Reports database – 180,000 messages/sec - ~ 50% increase in processing speed
For the same application, completion time decreased by 66%
Data Mart - daily transactions of 2 TB applied to standby in real-time
Data Loads – processing times decreased by 50%
Data Mart – hybrid columnar compression - 4x compression gain vs ACO. 15-16x compression achieved on raw data using HCC
Full Backup time of a ~150 TB data mart reduced by 85% (over Infiniband switches)
Utilize SmartScan capabilities
Migration
November 17, 2015 51
Consolidation planning
Database load compatibility analysis setup appropriate resources – parameters, limits
Migration to Exadata using Data Guard, RMAN Closing and RMAN Duplicate
Plan hardware - Dedicated network
Test it out – connectivity, test with application and RAT testing for performance validation
CONSIDERATIONS
November 17, 2015 52
Coordinate downtime
Risk Mitigation: Manage capacity Oracle Homes – 2 Load Isolation Instance caging, IORM deployed to manage resources consolidated
Security isolation
Rolling upgrades of databases
Rolling Exadata bundle patches – cell servers, components, compute nodes and clusterware
Observe utilization Re-evaluate database parameters
Reduced time to market
Platform as a service
Database as a service
Reduced cost Hardware, licensing operations
BENEFITS OF CONSOLIDATION
Copyright © 2015, Oracle and/or its affiliates. All rights reserved. 54