4. How Did We Get Here?
Exadata V1 - September 2008
•
HP Hardware / Data Warehouse only
•
Saw it, decided to wait
Exadata V2 - September 2009
•
Local Customer Purchased V2 Full Rack
•
Working with customer, saw the awesome potential
•
Six months later, purchased our V2 Quarter Rack
Exadata X2 - September 2010
•
Upgraded our V2 Quarter with an X2 Quarter
4
5. Proofs of Concept
POC’s are very popular
Results are often too good…
Exadata has an Acceptance Cycle:
•
Denial
•
Anger
•
Depression
•
Acceptance
5
6. How is Exadata being Used?
•
39 Companies
•
Range from Non-Rac to 16 Node RAC
•
Baby Exadata to X2-2 Full and X2-8 / X3-8 Full
•
More High Capacity Drives than High Performance
Data ERP Mixed
Consolidation Warehouse eBiz DW / OLTP OLTP
Peoplesoft
5% 40% 13% 33% 9%
Retail, Energy, Marketing, Healthcare, etc
6
7. Organizational Challenges
Who Should Manage The Beast
•
General Thinking is DBA’s (DMA’s ?)
•
Patching Requires More Knowledge Than Most DBAs Currently Have
-
Linux
-
Network
-
Hardware
-
Storage
•
Best Approach for Most is Combination of DBA / Sysadmin
•
SAN Guys are Out of the Picture
7
8. Top 4 reasons for Choosing Exadata
•
Consolidation
•
Performance
•
High Availability
•
Implementation Platform
-
Rapid Instantiation
-
Number of instances
-
Fast clones
8
9. Top Apps running on Exadata
•
Custom Data Warehouse / OLTP
•
PeopleSoft
•
EBS
•
OBI Apps
9
10. Alternatives Considered
•
Netezza
•
Teradata
•
Roll your own
-
Traditional Hardware / Storage
-
Solid state disk
-
InfiniBand
-
Flash cache
10
11. Common Configurations
•
¼ Rack – High capacity drives
•
Auto DOP – Off
•
Buffer Cache– Smaller than on non-Exadata
•
Flash – Popular tables pinned
•
Huge Pages – Enabled (no AMM)
•
Backups – RMAN to RECO then to tape
11
12. Common Migration Strategies
Logical
•
Data Pump
•
exp / imp
•
Golden Gate
•
CTAS Across DBLink
Physical
•
RMAN
•
Transportable Tablespaces
•
Dataguard Physical Standby
12
13. Typical Performance Results
Current Exadata Exadata Exadata Parallel Exadata
System 8G SGA 15G SGA 40G SGA Degree Improvement
8G SGA Factor (based on
8G SGA)
Query 1 46:13.00 00:00.02 00:00.02 00:00.02 24 138,650
Query 2 58:55.00 00:00.05 00:01.66 00:01.94 24 70,700
Query 3 32:24.00 11:47.44 10:29.40 08:20.10 24 3
Query 4 06:57.00 00:15.81 00:15.45 00:15.80 24 26
Query 5 8:45:12.00 13:17.68 10:36.32 11:05.40 24 40
Query 6 14:04.00 00:25.14 00:11.60 00:11.83 24 34
Query 7 04:47.00 00:16.46 00:16.80 00:18.97 24 17
Query 8 08:33.00 00:36.71 00:35.31 00:35.22 12 14
Query 9 6:38:10.00 02:50.14 02:49.07 02:48.65 Serial 140
Query 10 19:59.00 10:43.30 06:48.19 03:33.01 12 2
Improvement factors are based on the current system
Compared to the Exadata with an 8G SGA
13
14. Indexing Strategies
•
Single Row Access (OLTP) Needs Indexes
•
Challenge is to Use Indexes When
Appropriate
•
You Probably Need Fewer Indexes
•
You May Have to Get Creative
-
optimizer_use_invisible_indexes
-
db_multi_block_read_count
-
optimizer_index_cost_adj
-
system stats (Exadata mode) ???
•
Normal plan control mechanisms
14
15. Typical Compression Strategies
•
HCC Provides Exceptional Compression Ratios
-
10X is pretty good guess
-
6X – 60X in Practice
•
Oddly Enough Many are Not Using HCC
•
HCC Not Appropriate for Active Data
•
HCC Complements Partitioning
-
Requires Direct Path Loads
-
Update Move
-
Single Row Update Locks Entire CU
-
Falls Back to OLTP
15