Contenu connexe
Similaire à Epic Clarity Running on Exadata
Similaire à Epic Clarity Running on Exadata (20)
Epic Clarity Running on Exadata
- 1. Optimize the Performance of Your
Epic Clarity Data Warehouse
Industry specific cover image
Webcast 2/14/2013
||
| Epic
Anita Salinas
Patrick O’Connor
Tim Fox
Bob Bryla
Healthcare
Bus. Dev.
Oracle
Healthcare
Sales Consultant
Oracle
Chief Technologist
Enkitec
Snr. DB Architect &
Systems Engineer
Epic
© 2013 Oracle Corporation
- 2. Agenda
• Introductions
• Why optimize?
• Exadata: extreme performance for OLTP and DW
• Customer results: Enkitec
– Benchmark 1 results
– Benchmark 2 results
– Short demo
• Epic/Clarity target platforms explained: Epic
• Summary and next steps
• Q&A
© 2013 Oracle Corporation
2
- 3. Exadata Delivers Higher Value To Epic Clarity Users
Benefits Realized In Multiple Areas
IT Value
IT Cost Advantage
• Reduce core IT costs
• Significant cost benefits
• Lowest industry TCO
What if you could get
MORE information
SOONER
and USE LESS
hardware to do it?
© 2013 Oracle Corporation
•
•
•
•
•
Higher operational excellence, raise IT bar
Improve service - enhance SLA metrics
Seamless DW w/OLTP environment
Higher performance, scalability, throughput
Standardized complete management tools
Business Value
Epic
Clarity on
Oracle
Exadata
•
•
•
•
Improve quality of patient care
Receive timely critical reports
Execute reports more frequently as needed
Strategic partnership IT<->Business
3
- 4. Business Users Will Realize Significant Benefits
Oracle Customers Confirm Benefits
Epic Clarity Reports: 5-100x
Performance Improvements
ADT
Prelude
Sample Set of Reports
OpTime
Surgery
• Organ donor list, heart and lung transplant reports
• Specific inpatient diagnosis/flowsheet data related to transplants
EpicCare
Inpatient
• Currently admitted inpatient data for specific counties
EpixRx
Medication
• Medication, MAR, dispensed charge data
• Orders and treatment plan data
EpicCare
Outpatient
Tapestry
Resolute
Hospital
Billing
Profess.
Billing
• Orders, results, diagnosis for ambulatory visits for specific depts
• Outpatient appointment data for a specific county
• OR logs excluding specific CPT codes including charge data
• OR logs for specific CPT codes including charge data
• ED data from the prior day based on trauma diagnosis
• ED Order data
• ED patient flowsheet data and events
© 2013 Oracle Corporation
4
- 5. Customers Confirm Higher Business Value
Enhanced Patient Care With Confidence
~200 Daily Reports, >300
Locations Will Benefit
Timely Transplant Reports
Admin
• Improved patient care due to timely
information, data confidence
Other
Reports
• Enhanced productivity for all
coordinators, supporting personnel
• Improved IT productivity, eliminating
unnecessary running of reports
• Improved patient care
IT
Ops
• Significant productivity boost to clinical,
research, administrative users
• Improved operational effectiveness and
reduced cost to keep the lights on
Clinical
Finance
New Research Reports
• Meet new requirements due to
faster report execution
© 2013 Oracle Corporation
Research
Finance
Education
Provide financial reports to
analysts sooner for regular
reporting periods
5
- 7. Exadata Unified Workload Transformation
Single Machine for…
• OLTP
• Data Warehousing
• ETL
• Query parallelism
OLTP with Analytics and Parallelism of
Warehousing
Warehousing with Interactivity, Availability,
and Security of OLTP
© 2013 Oracle Corporation
7
- 8. Exadata Innovations
• Hybrid Columnar Compression
• Intelligent Storage
– Scale-out InfiniBand storage
– Smart Scan query offload
+
+
+
• Smart PCI Flash Cache
– Accelerates random I/O up to 30x
– Triples data scan rate
– 10x compression for warehouses
– 15x compression for archives
Data
remains
compressed
for scans
and in Flash
Benefits Cascade
to Copies
© 2013 Oracle Corporation
uncompressed
compress
primary DB
standby
test
dev
backup
8
- 9. Oracle Exadata: Extreme Performance and Scale
Advantages
• Significantly reduce query times by orders of magnitude
• Use fewer indexes to significantly improve daily load times
– Less space utilization
– Reduced maintenance of index builds/rebuilds
• Lower costs by consolidating all workloads on one platform
– Use Exadata for simultaneous Warehouse and OLTP
• Accelerate response times by up to 100x (or better)
© 2013 Oracle Corporation
9
- 10. Compression Ratio of Real-World Data
Query
Compression
Ratio
• Compression ratio varies by
customer and table
(Average=
13x)
Healthcare
C
Healthcare
B
• Trials were run on largest table at
10 ultra large companies
Financial
P
Financial
B
Financial
U
Financial
H
• Average revenue > $60 BB
• 13x – Avg query compression ratio
Telecom
A
Telecom
T
Telecom
H
• On top of Oracle’s already highly
efficient format
0
© 2013 Oracle Corporation
10
20
30
10
- 11. Secure Database Machine
•
Moves decryption from software to hardware
•
Over 5x faster
•
•
© 2013 Oracle Corporation
Near zero overhead for fully encrypted database
Queries decrypt data at hundreds of Gigabytes/
second
11
- 12. Epic Clarity on Exadata
Benchmark 1 Details
© 2012 Oracle Corporation
12
- 13. Observations - Epic Clarity on Exadata
• Data model has many very wide tables but rarely are all columns in a single report
• Data model loaded on daily / usually requires significant DB server resources
• Thousands of reports are run against Clarity on a daily basis
• Up to 120 reports may execute concurrently
• Clarity customers look for database configurations which improve throughput.
Often, the result is non-default Oracle configurations
• Customer-written report queries are often more complex than Epic-released
reports, and are challenging to tune with traditional methods
© 2013 Oracle Corporation
13
- 14. Epic Clarity on Exadata POC - Approach
• 1.5T Clarity database imported to Exadata X2-2 Quarter Rack (excluding audit tables)
• One BizObj server (VM) used to generate reporting load for 40 concurrent report jobs
• Evaluated automated reporting batches for execution time, load characteristics
• Customer supplied specific, long-running queries tested individually on Exadata
• Where applicable, Exadata features induced to explore performance
• Exadata’s Hybrid Columnar Compression (HCC) not used to compress tables during
the POC, but compression tests were run on large tables
• Tests on CLARITY_TDL_TRAN table show the following results
• Query High HCC Compression ratio – 8x to 10x
• Can reduce a 30GB table to 3GB
• Query Performance of HCC Compressed data often execute faster
© 2013 Oracle Corporation
14
- 15. Query Execution
• Customer supplied queries were executed under the following conditions:
• Database configured per Customer (matches current production)
• Reduced buffer cache to 2GB / multi-block read count = 128 / all non-PK indexes
made invisible
• Configuration changes were made to show that Exadata performs better, for
most DW workloads, with a smaller memory footprint
• The following page displays the results of the individual query testing done
for a Clarity customer on Enkitec’s Exadata X2-2 quarter rack
© 2013 Oracle Corporation
15
- 16. Results – Query Execution
Average Performance Improvement – 91x
© 2013 Oracle Corporation
16
- 17. Epic Clarity on Exadata
Benchmark 2 Details
© 2012 Oracle Corporation
17
- 18. Epic Clarity on Exadata POC – Approach
• Customer provided 2T production Clarity database export, 20 specific queries
• Supplied queries were run unmodified under three configurations:
*8 GB SGA (equivalent to current production) *15 GB SGA *40 GB SGA
• PARALLEL_MAX_SERVERS =24
• Used standard formula maximum parallel Servers = 2 * Core Count
•
•
•
•
•
•
Each query executed 2x to ensure at least some relevant data in buffer cache
Hybrid Columnar Compression (HCC) was not used
No tables were pinned in Exadata Smart Flash Cache
The entire POC was run on a single node Exadata Quarter Rack
Parallel slaves were confined to one node of the RAC
All serial processes were run on a single node of the RAC
© 2013 Oracle Corporation
18
- 19. Results – Query Execution
Currnent
System
8G SGA
Query 1
Query 2
Query 3
Query 4
Query 5
Query 6
Query 7
Query 8
Query 9
Query 10
46:13.00
58:55.00
32:24.00
06:57.00
8:45:12.00
14:04.00
04:47.00
08:33.00
6:38:10.00
19:59.00
Exadata
8G SGA
00:00.02
00:00.05
11:47.44
00:15.81
13:17.68
00:25.14
00:16.46
00:36.71
02:50.14
10:43.30
Exadata 15G
SGA
00:00.02
00:01.66
10:29.40
00:15.45
10:36.32
00:11.60
00:16.80
00:35.31
02:49.07
06:48.19
hr:min:sec:10th
sec
Exadata 40G
SGA
00:00.02
00:01.94
08:20.10
00:15.80
11:05.40
00:11.83
00:18.97
00:35.22
02:48.65
03:33.01
Parallel Degree
Exadata Improvement
Factor (based on 8G SGA)
24
24
24
24
24
24
24
12
Serial
12
138,650
70,700
3
26
40
34
17
14
140
2
Improvement factors are based on the current system compared to Exadata with an 8G SGA
© 2013 Oracle Corporation
19
- 20. Results – Query Execution Continued
Currnent
System
8G SGA
Query 11
Query 12
Query 13
Query 14
Query 15
Query 16
Query 17
Query 18
Query 19
Query 20
© 2013 Oracle Corporation
28:07
40:07
36:08
1:10:27
04:45
02:57
1:27:26
42:32
18:23
3:13:31
Exadata
8G SGA
00:13.18
01:58.41
00:12.15
09:29.83
00:13.68
01:33.33
08:05.57
02:24.66
00:05.03
00:18.39
Exadata 15G
SGA
00:13.66
01:52.75
00:13.82
03:25.33
00:13.80
00:00.46
00:13.49
01:21.20
00:14.04
00:15.75
Exadata 40G
SGA
00:14.24
01:55.74
00:11.96
00:13.52
00:13.37
00:02.14
00:13.32
00:58.96
00:13.76
00:16.67
Parallel Degree
Exadata Improvement
Factor (based on 8G SGA)
24
24
24
Serial
24
24
Serial
24
24
24
128
20
178
7
21
2
11
18
219
631
20
- 21. Results – HCC Compression Test
To test Hybrid Columnar Compression on Clarity data, the Compression
Advisor (DBMS_COMPRESSION) was used to simulate compression of
the CLARITY_TDL_TRAN table
HCC Compression Level
Compression Ratio
Query Low
Query High
6 to 1
Archive Low
8 to 1
Archive High
© 2013 Oracle Corporation
3 to 1
10 to 1
21
- 23. Conclusions
1• Epic Clarity workload hits the sweet spot for Exadata
– Large data volume, long running queries
2• It is impossible to match Exadata’s IO capability for large table scans with any
other Oracle-capable platform
3• Additional benefits are available
– Hybrid Columnar Compression, Exadata Flash, and Parallelism
4• With minimal effort, Customer can identify the business benefit of extreme
performance gains shown during this POC
5• Exadata supports improved performance with smaller memory
– More databases can be run on same hardware vs. custom built systems
© 2013 Oracle Corporation
23
- 24. Epic Clarity Target Platforms
Epic
• Target platform definition
• Supported platforms
• Customer demand
• Industry trends
• Exadata in-house at Epic
© 2013 Oracle Corporation
24
- 26. Summary
What can YOU do generating MORE reports FASTER on LESS hardware?
• Extreme Epic Clarity performance on Exadata
– Up to 100x faster
• Do more (reports) with less (hardware) in less (time)
– 512 reports in 12 hours vs. 1604 reports in 4 hours
– 3x # of reports completed in ¼ the time
– Lower costs, consolidate workloads on same hardware
• Improve care quality
– More timely = better intelligence
– Actionable data at your fingertips sooner and/or more often
© 2013 Oracle Corporation
26
- 27. Next Steps
Join us at HIMSS13!
• Oracle and Enkitec Breakfast Briefing
Wed, March 6, 2013 7:30-9:00am
Register here
• Continue the Conversation Reception
Wed, March 6, 2013 4.30-7.30pm
Invite forthcoming
Investigate further
– Exadata website
– Schedule a private consultation
© 2013 Oracle Corporation
Consultation
• Assess performance of Epic Clarity DW
• Review reports and queries to identify
opportunities that improve reporting
• Compare system to benchmark results
• Written performance recommendations
Contact info@enkitec.com
27
- 31. Appendix – Query Execution
Current
Customer
System
8488_sec_aun8fmpug9jk4
8207_sec_1vauja2xan534
6881_sec_232b9Czqbnn9
6833_sec_18mgrhn25hvk8
6827_sec_facj6p8f68drf
6820_sec_azgu4cxwvub3n
5890_sec_57rgm8v0jzpc1
5695_sec_5a02q7wg0k05x
5546_sec_at3uwh0bmvygv
03:46:17.33
06:14:16.34
06:06:15.45
02:53:32.03
01:40:23.40
00:27:34.90
00:31:11.20
00:50:19.90
00:49:20.10
Exadata per
Customer
16GB Buffer
00:55:41.50
01:23:19.67
01:22:36.91
00:49:11.32
00:28:55.56
00:10:30.08
00:04:25.41
00:25:42.47
00:06:56.32
Exadata per
Enkitec
4GB Buffer
00:50:10.97
01:06:36.11
01:05:02.66
00:30:56.22
00:25:10.01
00:01:52.60
00:13:44.35
00:23:57.29
00:07:38.94
Exadata per
Enkitec
2GB Buffer
01:08:15.80
01:19:04.93
01:15:24.70
00:35:38.72
00:26:51.30
00:02:05.00
00:12:29.06
00:23:35.28
00:07:03.25
Exadata No
Indexes
2GB Buffer
00:22:08.46
00:52:50.74
00:52:53.65
00:18:15.83
00:11:50.95
00:00:38.90
00:03:00.75
00:00:46.47
00:07:13.97
All queries improved in performance on Exadata with no tuning. No parallelism
was used. All queries were run on one node of the two node RAC.
© 2013 Oracle Corporation
31
- 32. Appendix – Query Execution
Current
Customer
System
5282_sec_13w3x29huvpzs
4742_sec_g6hmtqdhggcs7
4736_sec_1jkjps3basyz7
4728_sec_9fy866srqj1hz
4716_sec_1wuj2pzmdf0wk
4120_sec_3vu8b5sfmr8r6
3534_sec_fv2hr8d15q4tr
3383_sec_dvztmf02uqcya
3184_sec_gg5jrs56h19t2
3182_sec_gazv5xbhh0w5s
00:38:40.30
00:31:12.80
00:16:34.40
00:28:07.20
00:47:45.80
14:35:44.65
00:09:22.60
00:12:54.30
00:36:13.60
00:08:08.50
Exadata per
Customer
16GB Buffer
00:04:55.45
00:04:55.45
Killed
00:33:24.63
00:11:31.62
TEMP
00:11:08.64
00:00:09
00:00:00.52
00:00:52.88
Exadata per
Enkitec
4GB Buffer
00:05:05.03
00:00:01.28
Killed
00:23:31.59
00:06:15.23
TEMP
00:09:12.01
00:00:11.52
00:00:03.63
00:02:38.70
Exadata per
Enkitec
2GB Buffer
00:05:05.76
00:00:00.95
Killed
00:24:51.67
00:04:59.86
TEMP
00:09:36.35
00:00:04.15
00:00:03.52
00:02:18.52
Exadata No
Indexes
00:12:46.55
00:00:02.20
00:17:52.99
00:09:54.17
00:10:29.17
TEMP
00:00:33.87
00:00:04.57
00:00:07.08
00:01:33.66
All queries improved in performance on Exadata with no tuning with the
exception of two queries, both of which experienced plan digression due to
database version change.
© 2013 Oracle Corporation
32
- 33. Appendix – Additional Tuning
• Query 4736 ran for 16 minutes at Customer. Due to execution plan changes from
10g to 11g, the query never finished on Exadata.
• After removing all non-PK indexes, Query 4736 finished in 17 minutes on Exadata
(1 minute longer than on Customer production).
• The largest table in the query was still using a PK index. After removing this index
(via hint) the query ran in 3 minutes 42 seconds on Exadata (5x faster).
© 2013 Oracle Corporation
33